analysis and interpretation of lower limb ... · emg activity was >75% for all muscles sites...

100
Analysis and Interpretation of Lower Limb Mechanomyographic Signals in Human Gait by Katherine Plewa A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy Institute of Biomaterials and Biomedical Engineering University of Toronto © Copyright by Katherine Plewa 2018

Upload: others

Post on 26-Jul-2020

0 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

Analysis and Interpretation of Lower Limb Mechanomyographic Signals in Human Gait

by

Katherine Plewa

A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy

Institute of Biomaterials and Biomedical Engineering University of Toronto

© Copyright by Katherine Plewa 2018

Page 2: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

ii

Abstract

Analysis and Interpretation of Lower Limb

Mechanomyographic Signals in Human Gait

Katherine Plewa

Doctor of Philosophy in Engineering

Institute of Biomaterials and Biomedical Engineering

University of Toronto

2018

This thesis investigates the patterns of accelerometer-based MMG in the lower extremities during

gait. To characterize dynamic muscle activity, we measured lower limb mechanomyography

(MMG) and electromyography (EMG) during over ground and treadmill locomotion in typically

developing youth and adults. First, MMG activity was validated against coincidental EMG activity

and detection parameters were optimized for each muscle using a particle swarm optimization

(PSO) algorithm. The mean balanced accuracy between MMG muscle activity and concurrent

EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity

was only observed about 55% of the gait cycle. These findings suggest that during dynamic

movements, electrical muscle activity is not directly followed by mechanical activity of the muscle

fascicles since some muscles, like the gastrocnemius, contract isometrically and lengthen in the

absence of electrical activity. To understand how mechanical activity is coordinated between the

lower limb muscles, mechanical synergies were then extracted using non-negative matrix

factorization (NMF) analysis and compared against neural synergies. For treadmill walking and

running, mechanical muscle activity yielded lower dimensional synergies than did corresponding

electrical activity. Furthermore, mechanical synergies contained overwhelming co-activity of

muscles, possibly reflecting the active and passive lengthening of muscle fascicles as joints move

Page 3: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

iii

to complete a motor task. These synergies can be used to track therapy progression by analyzing

how mechanical patterns change with MMG biofeedback.

Based on MMG patterns observed during typical gait, a Smartphone application, GaitTool App,

was developed as an at-home gait therapy tool. MMG was recorded at the tibialis anterior and the

lateral gastrocnemius using wearable sensors that transmit packets of MMG data to an Android

device via Bluetooth. MMG data were analyzed for spatiotemporal features and triggered auditory

feedback in the form of a cadence-driven rhythmic auditory stimulus to allow for anticipatory

movement preparation and execution. The sonification of the alignment of MMG peaks provided

an aural cue to differentiate between typical and atypical gait patterns.

Page 4: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

iv

Acknowledgments

I’m so lucky to have such awe-inspiring people in my life who have provided me with many rich

opportunities to learn and have enabled me to experience so much of what this wonderful world

has to offer. I never imagined that life would ever lead me to this point, but I’m excited with the

path I’ve chosen, for the failures I’ve endured and the successes I’ve triumphed, and all the

incredible people who have shared their stories with me along the way. What started as an

academic endeavor quickly turned into the toughest, most personal exploration I’ve ever done –

I’m truly grateful for all the lessons I’ve learned along the way and I’m proud of the person I’ve

become.

I’d like to extend my deepest thanks to my supervisor, Dr. Tom Chau, for serving as both a rock

and a guiding light on this, often tumultuous, journey. Thank you for listening to and understanding

my struggles, thank your patience when I didn’t think I had it in me to continue, and thank you for

building me up when I had little faith in myself and in MMG. Your vision for a future filled with

endless possibilities is infectious, and I’m honored to have been able to work with you. You have

taught me that there is no challenge too big, and no person too small to make a difference in the

world.

I’d like to thank my committee members: Dr. Virginia Wright, Dr. Kei Masani, Dr. Michael Thaut,

and Dr. Lee Bartel. Thank you for asking tough and thoughtful questions that extended me past

my comfort zone. Your expertise and your ideas have been critical over the years, and I’m grateful

to be able to collaborate with truly groundbreaking researchers. I hope that our paths will cross

again in the future!

Thank you to the Postdocs: Dr. Ali Samadani and Dr. Silvia Orlandi. To Ali – thank you for

pushing me to think more like an engineer, for challenging me to work on my programming, and

for always making me laugh. To Silvia, your hard work and perseverance has given me the spark

that I’ve been looking for – thank you for pushing through the hard times, and for giving me the

reality checks to keep me in line. You are both exceptional researchers with brains so full of

knowledge and I’m lucky to call you my mentors and my friends.

Page 5: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

v

Thank you to my summer students: Matthew Patel, Olivia Paserin, and Matthew Silverman. Thank

you for your hard work and contributions to my thesis work. I hope you learned just as much from

me as I learned from you.

Thank you to all my PRISM lab mates, I’ve had a wonderful time learning alongside such bright

minds. Thank you for enduring my MMG struggles along the way, for letting me have my “mom

moments,” and for showing me that I really suck at board games. Thank you, to Ka Lun Tam and

Pierre Duez, for your patience and ongoing Matlab support, I’d probably be still coding in a circle

without you. A special thanks to Marcela Correa Villada, for understanding how frustrating MMG

is, and for always having faith in me to keep looking for answers – we’re blessed that God brought

us together for this. Thank you to my dear Zahra Emami, for being such a brilliant mind, a brilliant

heart, and a brilliant soul. You are soft yet fierce, a unique combination of fire and ice, and I’m so

lucky for all the pep talks, all the cries, and for all the times that you helped me see the best in

others and in myself. Thank you to Dr. Helia Mohammadi, you are one of the toughest lady bosses

I know. And finally, my warmest thanks to Dr. Amanda Fleury – thank you for being my person

on this adventure. Thank you for enduring the highs and the lows and for growing with me, no

matter the distance – I’m so grateful to have such a wonderful and badass friend in my life.

Thank you to all my friends, especially Michelle Gu and Sophie Wang, who have been there for

me through the good, the bad, and the really bad times…Real talk, I don’t know what I would have

done without you, you are my anchors. Thank you for all the laughs and all the tears, and all the

liters of wine consumed along the way – I’m lucky to have found friends who love each other like

family.

I’d like to thank my parents – for teaching me about hard work through example, for showing me

what it means to work as a team, and for teaching me to never give up on myself. Thank you for

your unconditional support and love, and for always setting the bar a little higher – in school and

in life. Thank you for challenging me to be the best human I can be.

Finally, my deepest thanks to Pauly, thank you for serving as my daily inspiration and my

motivation. I don’t know if you know how much respect I have for you – you’re the one who

pushes the limits and doesn’t take no for an answer. Thank you for doing all that you do. This

one’s for you, kid.

Page 6: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

vi

Table of Contents

Acknowledgments.......................................................................................................................... iv

Table of Contents ........................................................................................................................... vi

List of Abbreviations .......................................................................................................................x

List of Tables ................................................................................................................................. xi

List of Figures ............................................................................................................................... xii

Chapter 1 ..........................................................................................................................................1

Introduction .................................................................................................................................1

1.1 Motivation ............................................................................................................................1

1.2 Research Questions and Objectives .....................................................................................3

1.3 Thesis roadmap ....................................................................................................................4

Chapter 2 ..........................................................................................................................................6

A Novel Approach to Automatically Quantify the Level of Coincident Activity Between

EMG and MMG Signals .............................................................................................................6

2.1 Abstract ................................................................................................................................6

2.2 Introduction ..........................................................................................................................7

2.3 Methodology ........................................................................................................................8

2.3.1 EMG and MMG Signals ..........................................................................................8

2.3.2 Detecting Concurrent EMG and MMG Activity .....................................................9

2.3.3 Muscle-Specific Optimization of Intermodal Agreement ......................................11

2.3.4 K-fold Cross-Validation.........................................................................................12

2.3.5 Statistical Analysis .................................................................................................12

2.4 Results ................................................................................................................................13

2.4.1 Muscle-Specific Detection .....................................................................................14

2.5 Discussion ..........................................................................................................................15

2.5.1 Criterion-Driven Quantification of EMG and MMG Agreement ..........................15

Page 7: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

vii

2.5.2 Need for Parameter Optimization ..........................................................................16

2.5.3 General Agreement Between MMG and EMG Activation During Gait ...............17

2.5.4 Discrepancies Between EMG and MMG for the Gastrocnemius Muscle .............18

2.6 Conclusions ........................................................................................................................20

Chapter 3 ........................................................................................................................................21

Comparing Electro- and Mechano-myographic Muscle Activation Patterns in Self-Paced

Pediatric Gait .............................................................................................................................21

3.1 Abstract ..............................................................................................................................21

3.2 Introduction ........................................................................................................................23

3.3 Methodology ......................................................................................................................24

3.3.1 Participants .............................................................................................................24

3.3.2 Instrumentation ......................................................................................................25

3.3.3 Data Collection ......................................................................................................26

3.3.4 Signal Processing ...................................................................................................26

3.3.5 Co-incident EMG-MMG Activity .........................................................................27

3.3.6 MMG Stride Characterization................................................................................28

3.4 Results ................................................................................................................................29

3.5 Discussion ..........................................................................................................................32

3.5.1 Coincident MMG and EMG Activity ....................................................................32

3.5.2 Discrepant MMG and EMG Activity ....................................................................33

3.5.3 Distribution of MMG Signal Power Over Gait Cycle ...........................................33

3.5.4 Differences Between MMG and EMG Signal Power Distribution Over the

Gait Cycle ..............................................................................................................34

3.5.5 Limitations and Future Work .................................................................................35

3.6 Conclusions ........................................................................................................................35

3.7 Acknowledgements ............................................................................................................36

Chapter 4 ........................................................................................................................................37

Page 8: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

viii

Mechanical Synergies During Gait as Revealed Through Mechanomyography ......................37

4.1 Abstract ..............................................................................................................................37

4.2 Introduction ........................................................................................................................38

4.3 Methodology ......................................................................................................................39

4.3.1 Participants .............................................................................................................39

4.3.2 Data Collection and Instrumentation .....................................................................40

4.3.3 Experimental Setup ................................................................................................40

4.3.4 Data Pre-Processing ...............................................................................................40

4.3.5 Muscle Synergy Extraction ....................................................................................41

4.3.6 Walking vs. Running .............................................................................................42

4.4 Results ................................................................................................................................43

4.4.1 Electro-mechanical muscle activity in gait ............................................................43

4.4.2 Extracting Muscle Synergies .................................................................................44

4.4.3 Neural Synergies ....................................................................................................46

4.4.4 Mechanical Synergies ............................................................................................47

4.4.5 Walk vs. Run ..........................................................................................................49

4.4.6 Reconstruction of Muscle Signals .........................................................................50

4.5 Discussion ..........................................................................................................................52

4.5.1 Electromechanical Activity during Gait ................................................................52

4.5.2 Muscle Synergy Analysis ......................................................................................53

4.6 Conclusions ........................................................................................................................56

4.7 Acknowledgments..............................................................................................................56

Chapter 5 ........................................................................................................................................57

Designing a Wearable MMG-based Mobile App for Gait Rehab.............................................57

5.1 Abstract ..............................................................................................................................57

5.2 Introduction ........................................................................................................................58

Page 9: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

ix

5.3 Mobile App Design ............................................................................................................59

5.3.1 Design Considerations ...........................................................................................60

5.3.2 MMG Muscle Activity ...........................................................................................61

5.3.3 Arduino Processing ................................................................................................62

5.3.4 Gait Analysis and Feature Extraction ....................................................................63

5.3.5 Auditory Biofeedback ............................................................................................63

5.3.6 Use Case.................................................................................................................64

5.4 Mobile App Implementation ..............................................................................................65

5.5 App Evaluation ..................................................................................................................66

5.6 Discussion and Conclusions ..............................................................................................66

5.7 Acknowledgments..............................................................................................................67

Chapter 6 ........................................................................................................................................68

Conclusions ...............................................................................................................................68

6.1 Summary of Contributions .................................................................................................68

6.2 Future Work .......................................................................................................................69

6.3 Publications ........................................................................................................................70

6.3.1 Journal Articles ......................................................................................................70

6.3.2 Conference Presentations .......................................................................................70

References ......................................................................................................................................72

Page 10: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

x

List of Abbreviations

CNS Central nervous system

RAS Rhythmic auditory stimulus

CP Cerebral palsy

MMG Mechanomyography

EMG Electromyography

TA Tibialis anterior

LG Lateral gastrocnemius

MG Medial gastrocnemius

VL Vastus lateralis

BF Biceps femoris

TP True positive

FP False positive

TN True negative

FN False negative

𝜏 Amplitude threshold

𝜔 Moving window size (ms)

𝜔 Activity overlap (%)

BACC Balanced accuracy

PSO Particle swarm optimization

GLM General linear model

AMG Acoustic myography

RMS Root-mean-square

FSR Force sensitive resistor

MTC Musculotendinous complex

NMF Non-negative matrix factorization

VAF Variability accounted for

𝑀 Muscle activation pattern

𝑊 Muscle synergy weights

𝐶 Temporal activation coefficient

SW Slow walk (3 mph)

FW Fast walk (4 mph)

SR Slow run (5 mph)

FR Fast run (6 mph)

W Grouped Walk

R Grouped Run

W+R Global Walk + Run

Page 11: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

xi

List of Tables

Table 3-1 Maximums co-incident activity between EMG-MMG signals by muscle and side (right

vs. left leg) as measured by maximum balanced accuracies (first row). The subsequent rows

report the corresponding optimal values of window sizes and amplitude thresholds. Values

shown are mean and standard deviation across all participants. ................................................... 29

Page 12: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

xii

List of Figures

Figure 1-1 Thesis roadmap ............................................................................................................. 5

Figure 2-1 - Determining coincident activity of EMG (blue) and MMG (red line) signals using an

amplitude threshold (black dotted line) and moving window (depicted by the boxes). Examples

of true negative (TN), true positive (TP), false negative (FN) and false positive (FP) cases, along

with the three parameters, the amplitude threshold (τ), the moving window size (ω), and the

minimum percent of EMG-MMG activity overlap (δ), are illustrated. ....................................... 10

Figure 2-2 Boxplot of the averaged balanced accuracy across participants at each muscle for the

left (yellow) and right (blue) sides. The central mark denotes the median for each muscle, the

edges of the box indicate the 1st and 3rd quartiles, whereas the whiskers denote extremes in the

data and the ‘+’ symbols represent outliers. ................................................................................. 14

Figure 2-3 Mean optimized PSO parameters across participants - the amplitude threshold (τ), the

moving window size (ω), and the minimum percent of EMG-MMG activity overlap (δ) are

shown for each sensor on the left (yellow) and right (blue). The boxplot edges represent the 1st

and 3rd quartiles with the central line representing the median, and the whiskers denote extremes

with the ‘+’ showing outliers in the data. ..................................................................................... 15

Figure 2-4 Example of one participant's optimized segmentation results for the right tibialis

anterior showing detection results corresponding to: (a) true positive (TP), (b) true negative

(TN), and (c) false positive (FP). For this participant (P16) and sensor, we observed an

intermodal agreement of 96%. ...................................................................................................... 18

Figure 3-1 EMG and MMG sensors attached to the participants' muscles (TA = tibialis anterior,

LG = lateral gastrocnemius, VL = vastus lateralis, BF = biceps femoris) shown on the left leg,

and the backpack worn by the participant containing the MMG data board and tablet (right). ... 26

Figure 3-2 Determining coincident activity of EMG and MMG signals using an amplitude

threshold (black line) and moving window (box). Examples of true negative (TN), true positive

(TP), false negative (FN), and false positive (FP) cases are illustrated. ....................................... 28

Page 13: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

xiii

Figure 3-3 Balanced accuracies (BACC) averaged across all participants, showing degree of

EMG and MMG signal alignment as a function of amplitude threshold (vertical axis) and

window size (horizontal axis). BACC closer to 1 indicates greater coincident activity between

EMG and MMG. ........................................................................................................................... 30

Figure 3-4 The typical activity patterns are shown for one participant (P28), showing the mean

(black) with standard deviation (red) for all muscles of the right leg. The gait cycle begins at heel

strike and swing phase typically begins at 60% of the gait cycle. ................................................ 31

Figure 3-5 Mean EMG (clear box) and MMG (shaded box) signal power across all participants

for the right (top row) and left (bottom row) legs. The asterisks (*) specify significant differences

between EMG and MMG power within an interval of the gait cycle division. ............................ 32

Figure 4-1 An example of EMG and MMG during one gait cycle from the right leg recorded

during each of the treadmill speeds (slow walk (SW), fast walk (FW), slow run (SR), and fast run

(FR)) for a representative participant (P10). ................................................................................. 44

Figure 4-2 Scree plot of overall VAF (across participants) for each synergy level. Each row of

plots corresponds to VAF values for one condition: slow walk (SW), fast walk (FW), slow run

(SR), fast run (FR), walk (W), run (R), and global walk + run (W+R). ..................................... 46

Figure 4-3 An example of extracted neural synergies for the grouped conditions for one

representative participant (P08). At each synergy level, we show the corresponding synergy

weights (W) at each muscle (left (blue) and right (red) sides) and the mean synergy coefficients

(C) that together account for at least 95% of the reconstructed muscle signals. .......................... 47

Figure 4-4 Mechanical synergies extracted for all conditions for a representative participant

(P08). Muscles are grouped together with left (blue) and right (red) bars. .................................. 49

Figure 4-5 Cosine similarity matrix of EMG (left) and MMG (right) synergy weights for grouped

walk vs. run conditions. EMG synergies show distinct patterns between synergy levels (syn1,

syn2, syn3), whereas MMG synergies show a lot of similarity between synergies and conditions

(more red areas). ........................................................................................................................... 50

Page 14: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

xiv

Figure 4-6 - In the NMF analysis, each original muscle signal (dotted line) is reconstructed

(black line) based on the synergy weights and synergy coefficients (coloured lines) through the

gait cycle. Shown are the reconstructions for EMG for Global W+R for P01. ............................ 51

Figure 4-7 - Reconstructed MMG signals based on two synergy levels in the NMF analysis.

Shown is P01 based on the Global condition................................................................................ 52

Figure 5-1 System flow of GaitTool app showing the main components at both the user and

system levels: MMG muscle activity measurement (A), gait analysis and feature extraction (B),

and auditory biofeedback (C). ....................................................................................................... 60

Figure 5-2 MMG sensors taped directly onto the muscle bellies of the tibialis anterior (A) and

lateral gastrocnemius (B). In this initial prototype, the user is able to carry the Arduinos in his

pockets during gait. ....................................................................................................................... 61

Figure 5-3 Example of filtered MMG of TA (blue) and LG (orange) showing aligned peaks (left)

that create a harmony and misaligned peaks (right) that do not create a harmony. ...................... 62

Figure 5-4 Example of filtered MMG of TA (blue) and LG (orange) showing aligned peaks

(left) that create a harmony and misaligned peaks (right) that do not create a harmony. ............. 64

Page 15: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

1

Chapter 1

Introduction

1.1 Motivation

Neurological lesions, located in the brain or spinal cord, are characterized by an impairment of the

central and peripheral nervous systems leading to sensory and motor dysfunction. Central effects

primarily result in spastic paralysis, of which there are five main deficits: overreaction to

stretching, selective motor control reduction, return to primitive locomotor patterns, muscle phase

changes, and altered proprioception [1]. The integration of sensorimotor information is needed for

functional movements [2], and motor control theories reinforce that human movements are

governed by centrally activated motor programs and modulated by sensory inputs [3]. Therefore,

measuring impaired muscles during gait may provide a pathway for identifying motor patterns

contributing to errors in atypical movements.

Gait is a complex task to learn, with both a voluntary and automatic process that require interaction

between neural networks in the central and peripheral nervous systems with the connecting

musculature [4-6]. Gait patterns have been shown to alter with aging and neuromuscular disease

[5, 7, 8] and with the use of assistive gait devices [9]. Therefore, many rehabilitation efforts focus

on the plasticity of the spinal cord and nervous system in order to enhance sensory and motor

function. Gait training has shown use-dependent plasticity leading to functional recovery of step

patterns, and suggesting that clinical interventions focus on relearning functional movements [3,

4]. Pediatric therapy has shown that even if children are not working on a meaningful functional

activity, the repetitive and concentrated practice may be playing a role in neural plasticity [10].

Biofeedback therapy provides active information about real-time physiologic responses to

facilitate the acquisition of voluntary control over those responses; in this way, biofeedback may

provide the sensory stimulation needed to modify the CNS to regain normal patterns of movement

[10, 11]. Currently, there is a paucity of research on biofeedback training in pediatric neurological

rehabilitation.

Page 16: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

2

Neurological rehabilitation makes use of various treatment modalities, including techniques from

neurologic music therapy, which can promote motor learning in children with neurological deficits

through auditory entrainment [12, 13]. The link between auditory and motor systems is evident

when applying rhythmic entrainment to movement disorder rehabilitation – the firing rates of

motor neurons are triggered by music and entrained to the auditory rhythm, thus driving the motor

rhythm into different frequency levels [13, 14]. Studies providing fixed rhythmic auditory stimulus

(RAS) have shown improvements in gait patterns and stride parameters for patients with stroke,

Parkinson’s disorder, traumatic brain injury, and cerebral palsy [15-18]. In children with cerebral

palsy (CP), damage to the motor cortex disrupts normal processes for motor control thereby

affecting rhythmic movements. Studies have shown improvements to symmetry and stride rate

with both therapy-guided and self-guided RAS gait therapies in children with CP, suggesting the

need for at-home therapies [15]. Therefore, designing a gait intervention that is both wearable,

improves access to rehabilitation and the potential for increased quality of life, which encompasses

the child’s perception of their social, physical and emotional well-being that evolves through

development [19].

Mechanomyography (MMG) is a method for measuring muscle activation, typically using

accelerometers or microphones, and has recently been introduced as an effective biofeedback tool.

MMG is the mechanical equivalent to electromyography (EMG), and has been used to describe

motor control strategies, in terms of the number of active motor units and firing rates [20-24].

Moreover, presenting MMG biofeedback during computer work has been beneficial in reducing

muscle fatigue [25] and is more accurate than EMG at showing muscle recovery from low force

contractions [21, 26]. Additionally, the development of accelerometer-based wearable

technologies makes MMG an appealing low-cost modality for long-term gait monitoring [27].

However, MMG research has been limited in complex, dynamic motor tasks because MMG signals

are susceptible to motion artifact [28-30]. Although gait studies have incorporated MMG with

functional electrical stimulation [31-34], there have been no studies measuring the spatiotemporal

patterns of MMG during gait. This information may provide the missing afferent information

needed to enhance motor control and the restoration of healthy gait patterns. MMG has many

promising applications in rehabilitation [30, 35, 36], including motor recruitment and

neuroplasticity after an injury [37], assessing pain with movement [38], and clinically with the use

of assistive devices and prosthetics [39, 40].

Page 17: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

3

The overall objective of this thesis was to study lower limb MMG signal behavior during gait and

subsequently exploit the identified patterns in the development of a novel gait therapy tool for

environments of daily living.

1.2 Research Questions and Objectives

To determine the feasibility of harnessing MMG as a viable biofeedback signal during self-paced

gait, the following research questions were asked:

1. How can MMG activity be automatically detected such that the alignment between MMG

and EMG recordings is maximized at each lower limb muscle during self-paced gait?

Additionally, are there muscle-specific differences in the tuning of detection parameters?

2. What are the differences in signal power between EMG and MMG signals over the gait

cycle? Specifically, are modality-specific (i.e., EMG vs. MMG) and muscle-specific

differences observed over the gait cycle?

3. What are the underlying coordinated patterns of MMG signals during specific movements,

such as walking and running? Specifically, how many levels of mechanical synergies are

required to coordinate the lower limb muscles, and are there differences in the mechanical

synergies for walking and running?

Through this exploration, features of the MMG signals will be identified in order to generate

an instantaneous auditory feedback in a way that aurally distinguishes between typical and

atypical gait patterns. This exploration leads to the final research question:

4. How can MMG signal features from two muscles be used to generate instantaneous

auditory feedback in a way that aurally distinguishes between temporally aligned and

misaligned muscle activities?

To answer these questions, the immediate objectives of this thesis were:

1. To develop a method to detect MMG activity concurrent with EMG activity during gait

across lower limb muscles

2. To characterize MMG spatiotemporal activity during self-paced gait.

Page 18: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

4

3. To determine the coordination of MMG activity during walking and running gait

Using these findings, MMG spatiotemporal patterns were analyzed and

4. Developed into a prototype of a Smartphone application for at-home gait therapy that:

a. Presents users with live auditory biofeedback of their MMG activity

b. Tracks user’s MMG synergies as an indication of their therapy progress

1.3 Thesis roadmap

The organization of this thesis is summarized in Figure 1-1. To address the above objectives, two

studies were completed. The results of the first study are presented in Chapters 2 and 3 which focus

on objectives one and two, validating and characterizing MMG-based muscle activity from

multiple lower limb muscles against the gold standard of surface EMG. These data were collected

from typically developing youth during self-paced gait. The second study of this thesis is detailed

in Chapter 4. This chapter addresses the third objective of this thesis. These data were collected

from typically developed adults during treadmill walking and running. Chapter 5 presents the

development of the “GaitTool” Smartphone Application (the final objective of this work) based

on rhythmic auditory stimulation (RAS) and sonography of the MMG signals recorded in Study

1. Finally, the main contributions of this research and suggested areas for future work are

summarized in Chapter 6.

Page 19: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

5

Figure 1-1 Thesis roadmap

Page 20: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

6

Chapter 2

A Novel Approach to Automatically Quantify the Level of Coincident Activity Between EMG and MMG Signals

2.1 Abstract

Although previous studies have highlighted both similarities and differences between the timing

of electromyography (EMG) and mechanomyography (MMG) activities of muscles, there is no

method to systematically quantify the temporal alignment between corresponding EMG and MMG

signals. We propose a novel method to determine the level of coincident activity in quasi-periodic

MMG and EMG signals. The method optimizes 3 muscle-specific parameters: amplitude

threshold, window size and minimum percent of EMG and MMG overlap to maximize the

agreement (balanced accuracy) between electrical and mechanical signals. The method was

applied to bilaterally recorded EMG and MMG signals from 4 lower limb muscles per side of 25

pediatric participants during self-paced gait. Mean balanced accuracy exceeded 75% for all

muscles except the lateral gastrocnemius (LG), where EMG and MMG misalignment was notable

(56% balanced accuracy). The observed temporal discrepancy between EMG and MMG activities

of the LG muscle can be interpreted in terms of the energy-conserving interaction between the LG

muscle and tendon during the gait cycle, resulting in the nearly isometric contraction of the LG

during stance. The proposed method can be applied to the criterion-driven comparison of any two

sets of biomechanical signals.

Page 21: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

7

2.2 Introduction

The determination of muscle contraction onset and offset during dynamic activities (i.e., those that

involve more than an isolated concentric or eccentric contraction) is useful for the assessment of

motor control and learning [41], tracking rehabilitation progress [42], training myoelectric control

of prostheses [27, 43, 44], and informing the design of robotics or access technologies [39, 45, 46].

Muscle activity can be revealed by mechanomyography (MMG), which is a method for measuring

the lateral oscillations of muscle fibres during contraction [41-43]. MMG is considered the

mechanical counterpart of electromyography (EMG), which is the conventional method for

measuring muscle activity [43]. As such, several kinesiological and clinical studies have deployed

MMG as a complementary signal to EMG in detecting neuromuscular pathologies and in

controlling multifunction access devices [42, 43, 47-49]. Additionally, bimodal systems with

EMG-MMG have been suggested for the identification of electromechanical efficiency in atrophic

or diseased muscle [50].

The identification of muscle activity is a precursory step to the study of contraction timing or

waveform morphology [51, 52]. Previous research has shown that the onset of MMG activity

generally corresponds to the onset of EMG activity in voluntary isometric contractions [53, 54].

However, it has also been recognized that the corresponding electrical-mechanical relationship

may not always be straightforward. In a previous work, we reported non-causal interactions

between electrical and mechanical activities, where mechanical activity may be present in the

absence of contemporaneous electrical activity and vice versa, during dynamic movements, such

as gait [55].

Given the non-trivial relationship between EMG and MMG signals, it is challenging to objectively

establish the level of their concurrence, especially in terms of the timing of muscle contractions

within a long recording of quasi-periodic activity such as walking. Many algorithms have been

proposed for the automatic detection of onsets and offsets of muscle activity from a single time

series, including methods based on amplitude thresholding [39], statistical modeling of activity vs.

rest [56], and signal feature classification [57]. However, it is not clear how these detection

methods could be extended to two simultaneously recorded time series, one EMG and the other

MMG, while affording a measure of their temporal agreement, in terms of concurrent

Page 22: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

8

manifestations of muscle contractions. Simple cross-correlation is inadequate as the phase lag

between EMG and MMG is not necessarily static over time [55].

There is thus a need for a detection algorithm that can objectively quantify the concurrence

between MMG and EMG activities. Since MMG and EMG manifestations of disordered

movements may vary among individuals, the method also needs to be subject-specific. Here, we

propose a method to optimize the detection of MMG activity based on concurrent EMG activity

recordings. The proposed formulation deploys subject-specific window size, amplitude threshold,

and minimum EMG-MMG overlap to maximize the concurrence of detected EMG and MMG

activities at various lower limb muscles during self-paced gait. Furthermore, we compare detection

accuracy and optimized parameters across muscle sites to assess the need for muscle-specific

tuning of the detection parameters.

2.3 Methodology

2.3.1 EMG and MMG Signals

25 typically developing pediatric participants (7 males and 18 females; average height 159.8 cm ±

11 cm and average weight was 56.6 ± 17.6 kg) between the ages of 8 and 18 (mean 14 ± 3) years

were recruited. All participants provided written informed consent. The study protocol was

approved by the Research Ethics Boards of Holland Bloorview Kids Rehabilitation Hospital and

the University of Toronto. A subset of recordings from 20 of the above participants was previously

analyzed in [55].

Electrical and mechanical muscle activities were simultaneously measured from the muscle bellies

of the tibialis anterior, lateral gastrocnemius, vastus lateralis, biceps femoris muscles, bilaterally.

Sensors were attached to the skin with double-sided tape, with MMG sensors about 3 cm proximal

to EMG sensors. MMG was collected at 1 kHz using tri-axial accelerometers (ADXL337, Analog

Devices Inc, Norwood, MA) positioned so that the z-axis was perpendicular to the longitudinal

axis of the muscle. Surface EMG (Trigno by Delsys Inc, Boston, MA, USA) data were collected

wirelessly at 2 kHz. Participants walked continuously at a self-selected pace for 15 min, with shoes,

counter-clockwise around an indoor circuit (5m × 8m with 100-lb linoleum flooring).

Page 23: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

9

Five non-overlapping 2-minute partitions of data were extracted from each participant’s recording,

for subsequent segmentation analyses. Partitions were selected visually to ensure continuous

walking (i.e., no stops) with no signal artifacts (i.e., unusually large signal deflections). The z-

component of the accelerometer signal was bandpass filtered (5th order Butterworth filter) between

5-50 Hz and full-wave rectified [29]. EMG signals were bandpass filtered (4th order Butterworth

filter) between 30-500 Hz, full-wave rectified, down-sampled to 1 kHz [51], and smoothed using

a moving window average of 101 samples. EMG and MMG signals were rendered zero-mean and

amplitudes were normalized from 0 to 1. All data analysis was carried out via a custom-designed

Matlab program.

2.3.2 Detecting Concurrent EMG and MMG Activity

The alignment between EMG and MMG signals was examined in two steps. The first was to

separately identify active regions within EMG and MMG signals. This procedure comprised

threshold-based activity detection over unity-normalized signals and involved tuning for the

normalized amplitude threshold-parameter, 𝜏. At any instant in time, a muscle was considered

“active” when the corresponding normalized signal exceeded its modality (MMG or EMG),

muscle and subject-specific amplitude threshold. Adjacent activity segments were merged together

if they were less than 10 ms apart, whereas segments that were less than 100 ms in duration were

discarded as they were likely non-physiological artifacts [39]. This first step yielded binary (1 or

0) indicator signals for MMG and EMG.

The second step was to evaluate the alignment of an MMG segment with its EMG counterpart

using balanced accuracy. A moving window of length, 𝜔 (ms), was used to divide the EMG and

MMG indicator signals into non-overlapping segments. The percent of overlap between EMG and

MMG segments was defined as the fraction of time in which both indicator segments were in the

ON state (value of 1). The percent of overlap was compared against a minimum overlap, (%),

to determine the appropriate label for the segment. A true positive segment was one where

concurrent EMG and MMG activity was identified whereas a window was labeled as false negative

when an active EMG segment occurred with no corresponding MMG activity. A true negative was

tallied when both EMG and MMG activities were absent within a window of time, while a false

positive was an instance of active MMG with no concurrent EMG activity (Figure 2-1).

Page 24: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

10

Figure 2-1 - Determining coincident activity of EMG (blue) and MMG (red line) signals using

an amplitude threshold (black dotted line) and moving window (depicted by the boxes).

Examples of true negative (TN), true positive (TP), false negative (FN) and false positive (FP)

cases, along with the three parameters, the amplitude threshold (𝝉), the moving window size (𝝎),

and the minimum percent of EMG-MMG activity overlap (𝜹), are illustrated.

Page 25: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

11

The definitions of true positives (𝑇𝑃), true negatives (𝑇𝑁), false positives (𝐹𝑃), and false

negatives (𝐹𝑁) can be succinctly expressed as follows:

𝑻𝑷 = ∑ 𝑰[𝑨𝒊 > 𝜹]𝒊 , 𝐰𝐡𝐞𝐫𝐞 𝑨𝒊 =𝟏

𝝎∑ 𝑰[𝑬𝑴𝑮𝒍 > 𝝉 ⋀ 𝑴𝑴𝑮𝒍 > 𝝉]𝝎

𝒍=𝟏 , (1)

𝑻𝑵 = ∑ 𝑰[𝑩𝒊 > 𝜹]𝒊 , 𝐰𝐡𝐞𝐫𝐞 𝑩𝒊 =𝟏

𝝎∑ 𝑰[𝑬𝑴𝑮𝒍 < 𝝉 ⋀ 𝑴𝑴𝑮𝒍 < 𝝉]𝝎

𝒍=𝟏 , (2)

𝑭𝑷 = ∑ 𝑰[𝑪𝒊 > 𝜹]𝒊 , 𝐰𝐡𝐞𝐫𝐞 𝑪𝒊 =𝟏

𝝎∑ 𝑰[𝑬𝑴𝑮𝒍 < 𝝉 ⋀ 𝑴𝑴𝑮𝒍 > 𝝉]𝝎

𝒍=𝟏 , (3)

𝑭𝑵 = ∑ 𝑰[𝑫𝒊 > 𝜹]𝒊 , 𝐰𝐡𝐞𝐫𝐞 𝑫𝒊 =𝟏

𝝎∑ 𝑰[𝑬𝑴𝑮𝒍 > 𝝉 ⋀ 𝑴𝑴𝑮𝒍 < 𝝉]𝝎

𝒍=𝟏 , (4)

where 𝐼[𝑃] is the Iverson bracket whose value is 1 if P is true and 0, otherwise. In the above

equations, the subscript 𝑖 denotes the 𝑖𝑡ℎ segment, and 𝐸𝑀𝐺𝑙 and 𝑀𝑀𝐺𝑙 indicate the 𝑙𝑡ℎ EMG and

MMG observations within the 𝑖𝑡ℎsegment of length 𝜔, respectively. The symbol ˄ denotes the

logical ‘and’ operation.

In summary, the amount of concurrent EMG and MMG activity was determined by three

parameters: the amplitude threshold (𝝉), the moving window size (𝝎), and the minimum percent

of EMG-MMG activity overlap (𝛿). These 3 parameters are depicted in

Figure 2-1.

2.3.3 Muscle-Specific Optimization of Intermodal Agreement

Particle swarm optimization (PSO) [58] was used to find a combination of participant-specific

segmentation parameters yielding the highest balanced accuracy in aligning EMG-MMG activity

at each sensor location. To identify muscle-specific concurrent MMG and EMG activity, we found

the normalized signal amplitude threshold (𝜏), the window size (𝜔), and minimum percent of

EMG-MMG activity overlap (𝛿) that together maximized the agreement between the active

regions of the two signals:

Page 26: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

12

𝐚𝐫𝐠𝐦𝐚𝐱𝝉,𝝎,𝜹

𝑩𝑨𝑪𝑪 (𝑬𝑴𝑮, 𝑴𝑴𝑮|𝝉, 𝝎, 𝜹), (5)

where , , , and 𝐵𝐴𝐶𝐶 (𝐸𝑀𝐺, 𝑀𝑀𝐺|𝜏, 𝜔, 𝛿) is the

balanced accuracy of detecting concurrent EMG and MMG activity given the parameter set

{𝜏, 𝜔, 𝛿}. The balanced accuracy was defined as:

𝑩𝑨𝑪𝑪 = 𝟏

𝟐(𝑺𝒆𝒏𝒔𝒊𝒕𝒊𝒗𝒊𝒕𝒚 + 𝑺𝒑𝒆𝒄𝒊𝒇𝒊𝒄𝒊𝒕𝒚), (6)

where

𝑺𝒆𝒏𝒔𝒊𝒕𝒊𝒗𝒊𝒕𝒚 =𝑻𝑷

𝑻𝑷+𝑭𝑵, 𝑺𝒑𝒆𝒄𝒊𝒇𝒊𝒄𝒊𝒕𝒚 =

𝑻𝑵

𝑻𝑵+𝑭𝑷 (7)

The search for this optimal combination of parameters was performed over a discrete grid of

amplitude thresholds ([0.0, 0.15] in increments of 0.005), moving window sizes ([10, 1500] in

increments of 10), and minimum percent of EMG-MMG overlap ([0.7, 1.0] in increments of 0.05).

2.3.4 K-fold Cross-Validation

To evaluate the reliability of the optimized parameters for segmenting muscle contractions during

gait, a k-fold cross-validation was performed. In each fold, the segmentation parameters were

optimized using the training data (a 2-minute partition) and the resulting optimized parameters

were tested on the remaining four partitions. The overall performance of the proposed approach

was reported in terms of its average balanced accuracy over the test segments of the 5-fold cross-

validation.

2.3.5 Statistical Analysis

The distributions of average BACC were non-Gaussian according to the Shapiro Wilks Test for

normality and hence, non-parametric statistical tests were invoked. To test for a potential effect of

sensor location (i.e., muscle) on average balanced accuracy, a Kruskal-Wallis multivariate analysis

10 1500 0.7 1.0 0 0.15

Page 27: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

13

of variance with Tukey- Kramer’s post hoc test was applied to average balance accuracy values

from each side (i.e., right and left) independently, with P<0.05 denoting a significant effect

(RStudio for R, Boston, USA).

A univariate generalized linear model (GLM) was then used to test for any effect of sensor location

with the interaction of the optimized parameters on balanced accuracy. The model used considered

the averaged balanced accuracy (BACC) as the dependent variable, the optimized parameters

(𝜏, 𝜔, 𝛿) as covariates, and, muscle and side as fixed factors. A Gamma (reciprocal) link function

was deployed. Significant effects were indicated by P<0.05 (RStudio for R, Boston, USA).

2.4 Results

The averaged balanced accuracy was greater than 75% at all sensor locations except at the

gastrocnemius (Figure 2-2). The gastrocnemius location had the lowest averaged balanced

accuracies at 48% (right) and 59% (left), whereas the vastus lateralis exhibited the highest

averaged balanced accuracies at 85% (right) and 83% (left). The Kruskal-Wallis test revealed a

significant effect of sensor location on mean balanced accuracy (p<0.001). Post hoc testing

identified the gastrocnemius sensor locations as having significantly lower balanced accuracy than

all other muscle locations on the right (p<0.001) and left (p<0.05) sides.

Page 28: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

14

Figure 2-2 Boxplot of the averaged balanced accuracy across participants at each muscle for the

left (yellow) and right (blue) sides. The central mark denotes the median for each muscle, the

edges of the box indicate the 1st and 3rd quartiles, whereas the whiskers denote extremes in the

data and the ‘+’ symbols represent outliers.

2.4.1 Muscle-Specific Detection

The three optimized parameters for each participant and sensor location can be seen in Figure 2-

3. On average, the highest BACCs across all muscles were obtained with a combination of an

amplitude threshold in the neighbourhood of 0.11, a window size of approximately 440ms, and a

minimum EMG-MMG overlap of 85%. The GLM model that included the interaction among the

optimized parameters (𝜏, 𝜔, 𝛿) and the interaction between muscles and sides (muscles×sides) did

not identify any interaction effect of optimized parameters and side on BACC (p>0.05). However,

there was a significant interaction effect among the optimized parameters on BACC (p<0.001).

Page 29: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

15

Figure 2-3 Mean optimized PSO parameters across participants - the amplitude threshold (𝝉), the

moving window size (𝝎), and the minimum percent of EMG-MMG activity overlap (𝜹) are

shown for each sensor on the left (yellow) and right (blue). The boxplot edges represent the 1st

and 3rd quartiles with the central line representing the median, and the whiskers denote extremes

with the ‘+’ showing outliers in the data.

2.5 Discussion

2.5.1 Criterion-Driven Quantification of EMG and MMG Agreement

We proposed a novel method for the automatic quantification of the level of coincident activity in

EMG and MMG signals. In our earlier work [21], we used an exhaustive search to detect the best

combination of window size ( ) and amplitude threshold ( ) at a fixed percent overlap of EMG

and MMG activity. With EMG as the reference, MMG-based muscular contractions were detected

in a single 2-min trial with balanced accuracies between 88% and 94% for all muscles except the

Page 30: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

16

gastrocnemius. In this study, we formalized a criterion-driven, systematic approach to quantifying

EMG and MMG agreement while also allowing the minimum percent of EMG-MMG overlap to

vary. It is important to note that the proposed method is not limited to quantifying temporal overlap

of activity (Equation 5) but allows one to measure the agreement between signal modalities using

different criteria. For example, one could instead choose the agreement between spectral features

in EMG and MMG data as the optimization criterion. In such case, we might expect low BACC

across muscles as concentric and eccentric contractions are known to manifest in MMG spectra,

but not in EMG [59]. This flexibility allows researchers to systematically explore and objectively

quantify similarities and differences in EMG and MMG signals, as they pertain to clinically

relevant movements [60]. Moreover, the proposed method can be generally deployed in the

comparison of an unknown signal against a template signal (e.g., indicative of health). The

proposed method can thus indicate the degree to which an unknown signal departs from the

template [60]. In particular, comparing recorded MMG activity against normative templates using

the proposed approach may provide a quantitative assessment of passive elastic mechanisms

present in neurotypical walking [61].

2.5.2 Need for Parameter Optimization

Our method optimized spatiotemporal parameters for each muscle of each participant, for each

signal modality. Given that EMG and MMG signals were normalized within-modality, the

optimized threshold was thus effectively modality-specific. In isometric contractions, the

morphology of EMG amplitudes increase from baseline and plateau at maximum contraction, then

decrease back to baseline as the contraction ends [62]. In contrast, the typical MMG morphology

of an isometric contraction comprises a high peak-to-peak amplitude at the onset of contraction, a

plateau in amplitude during the isometric hold, and a lower peak-to-peak amplitude during the

relaxation phase of the contraction [63], suggesting the need for an amplitude threshold slightly

below the plateau value. Given these known morphological differences between EMG and MMG

signals for well-studied isometric contractions, modality-specific thresholds were invoked for

dynamic contractions of gait, where EMG-MMG correspondence is less studied.

A moving window between 370 ms and 530 ms was selected by the optimization algorithm for

concurrent assessment of electrical and mechanical activity during self-paced gait. If we consider

that the tibialis anterior is electrically active for about 50% of the gait cycle [64], the selected

Page 31: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

17

window sizes are in line with previously reported gait parameters for school children, namely that

a typical stride rate is between 700 ms and 1000 ms [65]. Note that different window lengths may

be selected when considering other types of functional movements occurring on different time

scales. For example, typical lower limb muscle events during running are usually 50–100 ms in

duration, suggesting the use of shorter windows [66] to detect active MMG segments during gross

limb movement [29], or to identify participant-specific features in pathological movements, which

are often highly variable [67]. Limiting detection to the duration of physiological events of interest

through parameter optimization can also reduce false positives [8].

Generally, when using the proposed method to compare a measurement against a gold standard (in

this case EMG), detection parameters need to be set (through optimization) to hone in on the

differences (e.g., temporal, spectral, amplitude) of interest (specified by the criterion function).

2.5.3 General Agreement Between MMG and EMG Activation During Gait

At all but one muscle site, balanced accuracies exceeded 75%, suggesting general temporal

alignment of supra-threshold EMG and MMG signal segments. These results corroborate previous

work showing that electrical and mechanical activity of the lower limb muscles is predominantly

spatiotemporally aligned during self-paced gait [55]. As action potentials send electrical stimuli to

the motor units to contract (i.e., giving rise to EMG), muscle fascicles shorten or lengthen resulting

in corresponding mechanical activity (i.e., MMG signal) in the tibialis anterior, vastus lateralis,

and biceps femoris muscles [68]. Indeed, coincident EMG-MMG activity has been previously

reported for isometric and isokinetic contractions of upper and lower limb muscles [69-71], as well

as for dynamic contractions of the quadriceps during cycling [30]. In a gesture interaction system

[14], coincident EMG and accelerometer-based (mechanomyographic) activity was found to yield

high gestural onset and offset detection accuracy using a Hidden Markov Model. Our finding of

general temporal agreement between EMG and MMG activation in lower limb muscles during

walking thus resonates with previous literature. An example of optimized segmentation results

showing agreement and disagreement between EMG and MMG can be seen in Figure 2-4. For this

participant and sensor, the balanced accuracy was 96% and we present cases of true positive

segmentation, true negative, and false positive cases.

Page 32: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

18

a) – TP

b) TN

c) FP

Figure 2-4 Example of one participant's optimized segmentation results for the right tibialis

anterior showing detection results corresponding to: (a) true positive (TP), (b) true negative

(TN), and (c) false positive (FP). For this participant (P16) and sensor, we observed an

intermodal agreement of 96%.

2.5.4 Discrepancies Between EMG and MMG for the Gastrocnemius Muscle

Contrary to the general trend, in the gastrocnemius, we found significantly lower (BACC of

approximately 55%) EMG-MMG agreement compared to that observed at other sensor locations.

This discrepancy indicates that during walking, the electrical and mechanical activities of the

gastrocnemius do not align for about half of the gait cycle. Although we previously identified a

similar discrepancy between EMG-MMG amplitudes at the gastrocnemius [55], this finding

departs from the general theme of electromechanical correspondence reported in the MMG

literature. For example, when examining EMG and acoustic myography (AMG) during stepping

exercises, Harrison et al. noted temporal alignment of electro-mechanical signals recorded from

the medial gastrocnemius [70], while Beck et al. reported electro-mechanical signal

correspondence in the quadriceps during cycling [30]. The discrepancy discovered in our study

suggests that the relationship between EMG-MMG activities may be more complex in gait. We

elaborate upon this relationship below, focusing on the gastrocnemius muscle.

Muscle architecture is important in determining a muscle’s mechanical function [72, 73].

Page 33: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

19

Specifically, the lengthening capacity of a muscle is defined by the arrangement of the muscle

fibres and the sarcomeres. For example, the gastrocnemius is a bipennate muscle, which has shorter

muscle fibers, thus resulting in shorter fascicle displacements in comparison to longer-fibred

parallel muscles [73]. Furthermore, muscles with higher pennation angles and shorter fibre lengths

exhibit larger increases in muscle thickness during contraction [74]. Since MMG is reflective of

the initial length and volume changes of muscles during contraction [22], muscle architecture may

influence the pattern of mechanical activity observed during gait. When studying muscle fascicle

lengthening during the contact phase of human locomotion, Ishikawa, et al. [75] found a

discrepancy between fascicle shortening and low EMG activity of the soleus muscle during the

push-off phase of ground contact during gait. Interestingly, they did not observe the same

discrepancy between electrical activity and fascicle shortening of the medial gastrocnemius,

indicating that the pattern of fascicle length change is different between muscles [75]. Likewise,

our findings show that coincident EMG-MMG activity is muscle and function-specific. Thus,

generalizations about electrical and mechanical concordance should not be made between muscle

groups or functional movements.

Studies examining the behaviour of the muscle-tendon complex during gait report that the

gastrocnemius appears to contract nearly isometrically [76-78]. Given that MMG signal power is

known to be low during isometric contractions [25], one would thus expect minimal MMG activity

in the gastrocnemius while there is heightened EMG activity, which resonates with our observation

of poor temporal correspondence (~50%) between EMG and MMG activity of the gastrocnemius.

More specifically, Fukunaga, et al. [77] showed that the medial gastrocnemius muscle maintains

near-constant length while active during walking (contracts nearly isometrically), generating

minimal power with minimal energy cost. The tendon stretches during stance and recoils from the

beginning of single support to toe-off to create elastic strain energy. Part of this energy is released

by the tendon upon recoil, and is dissipated through the gastrocnemius during push-off. This

energy would contribute to some of the MMG vibrational activity observed after toe-off in swing

phase, and prior to heel strike.

Moreover, the gastrocnemius undergoes a concentric-eccentric contraction in the swing phase of

gait [79]. MMG has been shown to be more sensitive than EMG to the type of contraction being

performed. This is evident when observing the frequency spectra of eccentric versus concentric

contractions [59]. Specifically, MMG root-mean-square (RMS) amplitude is lower during

Page 34: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

20

isometric than concentric and eccentric contractions, and electromechanical efficiency (MMG

RMS/EMG RMS) is highest during eccentric contraction [63]. Combined with the evolution of

contraction type over the gait cycle, electromechanical efficiency would help to explain the

observed MMG response. During early stance, a low level of EMG produces some MMG through

a concentric contraction associated with low electromechanical efficiency. During mid-stance

EMG increases, but MMG remains low as the muscle contracts isometrically with low

electromechanical efficiency. During the latter part of swing phase, EMG restarts causing a larger

MMG deflection through the eccentric contraction with the highest electromechanical efficiency.

Finally, MMG preceding this deflection may be attributable to the propagated vibration from

tendon recoil. In other words, it appears that the observed EMG-MMG discrepancy is principally

due to the fact that the LG muscle has a nearly isometric contraction period during stance.

Incidentally, the LG is minimally activated during quiet standing [80].

2.6 Conclusions

We have proposed a novel method for systematically quantifying the level of temporal alignment

between electrical and mechanical muscle activities from simultaneously recorded EMG and

MMG signals during pediatric gait. When applied to signals collected from 25 pediatric

participants, electro-mechanical alignment was observed in the tibialis anterior, vastus lateralis,

and biceps femoris but not in the lateral gastrocnemius. The observed temporal discrepancy

between EMG and MMG may be attributable in part to the unique behavior of the LG muscle-

tendon complex during the gait cycle and its corresponding time-varying electromechanical

efficiency. The proposed method can be extended to quantitatively compare any two sets of

biomechanical signals according to a defined criterion.

Page 35: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

21

Chapter 3

Comparing Electro- and Mechano-myographic Muscle Activation Patterns in Self-Paced Pediatric Gait

The entirety of this chapter is reproduced from the following manuscript: Plewa, Katherine, Ali

Samadani, and Tom Chau. "Comparing electro-and mechano-myographic muscle activation

patterns in self-paced pediatric gait." Journal of Electromyography and Kinesiology 36 (2017):

73-80.

This is an author-created, un-copyedited version of an article published in the Journal of

Electromyography and Kinesiology. Elsevier B.V. is not responsible for any errors or omissions

in this version of the manuscript or any version derived from it.

© 2017. Elsevier B.V. http://dx.doi.org/10.1016/j.jelekin.2017.07.002

3.1 Abstract

Electromyography (EMG) is the standard modality for measuring muscle activity. However, the

convenience and availability of low-cost accelerometer-based wearables makes

mechanomyography (MMG) an increasingly attractive alternative modality for clinical

applications. Literature to date has demonstrated a strong association between EMG and MMG

temporal alignment in isometric and isokinetic contractions. However, the EMG-MMG

relationship has not been studied in gait. In this study, the concurrence of EMG- and MMG-

detected contractions in the tibialis anterior, lateral gastrocnemius, vastus lateralis, and biceps

femoris muscles were investigated in children during self-paced gait. Furthermore, the distribution

of signal power over the gait cycle was statistically compared between EMG-MMG modalities.

With EMG as the reference, muscular contractions were detected based on MMG with balanced

accuracies between 88-94% for all muscles except the gastrocnemius. MMG signal power differed

Page 36: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

22

from that of EMG during certain phases of the gait cycle in all muscles except the biceps femoris.

These timing and power distribution differences between the two modalities may in part be related

to muscle fascicle length changes that are unique to muscle motion during gait. Our findings

suggest that the relationship between EMG and MMG appears to be more complex during gait

than in isometric and isokinetic contractions.

Page 37: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

23

3.2 Introduction

Knowledge of muscle activity during dynamic activities such as gait is useful for the elucidation

of motor control and learning strategies [41], assessment of motor disabilities, monitoring the

progress of rehabilitation, development of neuroprosthesis [27, 43], and informing clinical

decision-making [42]. In particular, a comprehensive picture of muscle function can be derived

from continuous in situ recordings of muscle activity, as enabled through wearable technologies,

or “wearables.” Wearables consist of small, unobtrusive sensors attached to the body or to clothing

that monitor physiological and behavioral signals over extended periods of time. For example,

wearables utilizing movement sensors have been helpful in measuring gait dynamics, joint

kinematics, and the effectiveness of at-home rehabilitation efforts [27, 81]. Furthermore, these

smaller-sized wearables would be most useful in studying the smaller muscles in a pediatric

population.

Surface electromyography (EMG) records electrophysiological impulses over the muscle belly

during muscle contraction, and is the gold standard for measuring muscle activity [43]. EMG

applications historically include clinical diagnosis and rehabilitation monitoring, and more

recently, postural biofeedback, and control of human machine interfaces, such as communication

devices and video games [27, 82, 83]. An alternative to measuring the electrical impulses at the

onset of muscle activity is to measure the mechanical force transmission of muscle fibres via

mechanomyography (MMG). MMG measures the lateral oscillations of active and passive parts

of the series elastic component of the musculotendinous unit using various transducers, such as

microphones or accelerometers [41-43]. MMG has been used to describe motor control strategies,

in terms of motor unit summation [23], firing pattern during fatigue [21], and force during

contractions [63].

Some studies have explored the temporal relationship between EMG and MMG in voluntary

isometric and isokinetic dynamic contractions, reporting electromechanical delays between 20-

125 ms during voluntary contractions [84, 85]. Additionally, lower limb studies in cycle ergometry

have reported overlapping bursts of activity for simultaneously recorded EMG and MMG [53].

Consequently, a number of kinesiological and clinical studies have deployed MMG as a

complementary signal to EMG in assessing postural balance and age-related muscle changes [42,

43]. While EMG signals are not known to exhibit differences between concentric and eccentric

Page 38: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

24

contractions, Jaskólska, et al. [86] reported contraction-specific MMG frequency and amplitude

responses in the upper extremity agonist and antagonist muscles. Perry-Rana, et al. [87] discovered

differences in the EMG and MMG responses among the vastus lateralis, rectus femoris and vastus

medialis during maximal eccentric contractions while in an earlier study Perry, et al. [88] hinted

at an association between the metabolic and MMG characteristics of muscular contraction.

Additionally, the relationship between EMG and MMG frequency spectral patterns is more

complex in voluntary than in electrically induced contractions [89, 90]. Collectively, these findings

suggest that the relationship between MMG and EMG signals during gait may not be

straightforward and that MMG may bear uniquely complementary information about the

underlying muscle activity. To our knowledge, the temporal patterns of lower limb MMG and their

correspondence to simultaneously recorded EMG activity have not been explored during gait.

The availability of accelerometer-based wearables makes MMG a feasible alternative to printed

electrodes for EMG wearables [91]. Furthermore, a systematic review examining the use of

wearable accelerometer-based technology for neurological populations in the community

highlighted the effectiveness of these technologies in distinguishing between typical and atypical

mobility patterns [81]. Additionally, the smaller size of the accelerometer-based MMG sensor is

particularly conducive to application in pediatric rehabilitation settings. Although some studies

have deployed MMG alongside EMG or functional electrical stimulation in neuromuscular

populations [32, 92, 93], there have been no studies specifically focusing on MMG patterns in

pediatric populations. In particular, the level of temporal concordance between EMG and MMG

signals of the lower limbs during self-paced gait in children remains unknown.

In this study, we measured MMG of the lower limb muscles simultaneously with EMG during

self-paced gait in a typically developing pediatric population to evaluate the temporal patterns of

MMG activity during the gait cycle relative to EMG, and gain insight into the corresponding

muscle-specific, accelerometer-based mechanical activity.

3.3 Methodology

3.3.1 Participants

We recruited 20 typically developing pediatric participants (5 males and 15 females) between the

ages of 8 and 18 (mean 14.3±3.2, range 8-18) years with an average height of 158.8±11.6 cm and

Page 39: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

25

an average weight of 55.5±19.2 kilograms. This age group was chosen to ensure that participants’

gait patterns had developed into mature, adult-like patterns [94]. Participants reported no known

neurological disorders, or musculoskeletal injuries (current or previous) limiting their walking.

Participants wore athletic shoes and shorts during the study. The Holland Bloorview hospital and

University of Toronto Research Ethics Boards approved the study protocol, and all participants

provided informed written consent.

3.3.2 Instrumentation

Wireless surface electromyography (EMG) sensors (Trigno by Delsys Inc, Boston, MA, USA)

were fastened with double-sided tape over the muscle belly of each of the following lower limb

muscles: tibialis anterior (TA), lateral gastrocnemius (LG), vastus lateralis (VL), and biceps

femoris (BF). MMG data were collected using tri-axial accelerometers (ADXL337, Analog

Devices Inc, Norwood, MA). The accelerometers were positioned on the muscle bellies of the

same muscles, usually 3 cm proximal to the EMG sensor (Figure 3-1). For participants who were

less than 145 cm in height and under 33 kg in weight, we situated the accelerometer 2 cm from the

EMG sensor, to ensure proximity to the muscle belly and to avoid the corresponding tendon. For

each muscle, the accelerometer was oriented such that the z-axis was perpendicular to the

longitudinal axis of the muscle. An ultra-thin, force-sensitive resistor (FSR) (FSR 406, Interlink

Electronics, Shenzhen, China) was inserted inside the participant’s shoe to record heel strike for

each step [95]. All data were collected bilaterally.

Each sensor was connected to a data logging system housed inside a backpack (total mass: 1.83

kg) worn by the participant. All data were acquired via a custom-made LabVIEW program at a

sampling rate of 1 kHz for MMG and 2 kHz for EMG.

Page 40: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

26

Figure 3-1 EMG and MMG sensors attached to the participants' muscles (TA = tibialis anterior,

LG = lateral gastrocnemius, VL = vastus lateralis, BF = biceps femoris) shown on the left leg,

and the backpack worn by the participant containing the MMG data board and tablet (right).

3.3.3 Data Collection

Prior to the walking trial, we collected 20 seconds of baseline MMG and EMG during quiet

standing. Participants were then instructed to walk continuously at a self-selected pace on an

obstacle-free, rectangular-shaped, well-lit indoor gym track (5 m × 8 m with linoleum flooring).

All participants walked counter-clockwise around the track continuously for 15-minutes.

Participants did not complain of fatigue or discomfort from wearing the sensors and backpack, and

all participants were able to complete the full 15-minutes of walking.

3.3.4 Signal Processing

To extract MMG content from the accelerometry signals, the z-component of the signal was

bandpass filtered (5th order Butterworth filter) between 5-50 Hz and full-wave rectified. EMG data

were bandpass filtered (4th order Butterworth filter) between 30-500 Hz, full-wave rectified, and

down-sampled to 1 kHz [51]. Both MMG and EMG signals were smoothed using a moving

window average of 101 samples and separately normalized to the interval [0,1]. All data analysis

Page 41: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

27

was carried out via a custom-designed MATLAB program. We extracted 2 minutes of data starting

at the 2-minute mark of each participant’s 15 minute recording for subsequent EMG-MMG

temporal validation analyses.

3.3.5 Co-incident EMG-MMG Activity

The percent of overlap between EMG and MMG within a given window of observation was

defined as the fraction of time in which normalized versions of both signals exceeded an amplitude

threshold (described below), i.e., were “active”. Windows within which the overlap between MMG

and EMG active regions was at least 80% were identified as segments of co-incident activation.

To identify muscle-specific co-incident MMG and EMG activity, we found the window size and

the signal amplitude threshold that together maximized the agreement between the active regions

of the two signals. In other words, using EMG activity as the reference, we determined a window

length-amplitude threshold pairing that maximized the balanced accuracy of detecting MMG

activity. The exhaustive search for this optimal pairing was performed over a discrete grid of

amplitude thresholds (0.05 – 0.5 in increments of 0.05) and moving window sizes (100 – 1500 ms

in increments of 50). Balanced accuracy was defined as the arithmetic mean of sensitivity (true

positive rate) and specificity (true negative rate). A true positive occurred when MMG activity was

present given corresponding EMG activity whereas a false negative indicated an active EMG

segment with no corresponding MMG activity. A true negative was tallied when both EMG and

MMG activities were absent within a window of time, while a false positive was an instance of

active MMG with no concurrent EMG activity (Figure 3-2). Balanced accuracies were compared

across muscles with a one-way analysis-of-variance using RStudio for R (Boston, USA). One-way

analyses-of-variance were also invoked to evaluate across-muscle differences in window length

and amplitude thresholds.

Page 42: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

28

Figure 3-2 Determining coincident activity of EMG and MMG signals using an amplitude

threshold (black line) and moving window (box). Examples of true negative (TN), true positive

(TP), false negative (FN), and false positive (FP) cases are illustrated.

3.3.6 MMG Stride Characterization

For the 15-minutes of self-paced gait, EMG and MMG signals were segmented based on the heel

strikes as identified in the FSR signals. Fifty steps from the beginning and end of the session were

discarded to account for familiarization and termination effects, respectively, and steps <400 ms

and >2000 ms were also discarded to account for missteps or stops during the trial. With this

procedure, no more than 15 steps were eliminated for a given participant. The length of each stride

was then normalized to 2 seconds, and the power of the EMG and MMG signals was separately

calculated within each 20% division of the gait cycle. Muscle-specific differences in signal power

between EMG and MMG signals over the gait cycle were assessed using a repeated measures two-

way analysis of variance with measurement modality (EMG; MMG) and gait phase (0-20; 20-40;

40-60; 60-80; 80-100%) as independent factors. Significant differences in power were further

investigated via post-hoc pairwise t-tests with a Bonferroni adjustment for multiple comparisons.

The statistical analysis was performed using RStudio. The above analyses (comparing balanced

accuracies across muscles and the effects of measurement modality and gait phase on signal power)

Page 43: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

29

were replicated in subgroups of gender, age (8-12 years and 13-18 years) and body size (<60 kg

and >60kg).

3.4 Results

Balanced accuracies averaged across participants for different window lengths and amplitude

thresholds are shown in Figure 3-3. High accuracy suggesting strong coincidence between EMG

and MMG activities was seen for different amplitude threshold-window size pairings for the TA,

VL, and BF muscle sites. However, the maximum balanced accuracy for the LG muscle sites of

both right and left legs were significantly lower than that of all other muscle sites (p<0.001),

indicating low coincidence between MMG and EMG-demarcated activity. This finding is

corroborated when examining the maximum balanced accuracy achievable for each muscle on

each side, as shown in Table 1. No significant differences were seen among muscle sites for

window sizes (p =0.74) or threshold amplitudes (p = 0.48) for maximum balanced accuracies

across participants.

Table 3-1 Maximums co-incident activity between EMG-MMG signals by muscle and side

(right vs. left leg) as measured by maximum balanced accuracies (first row). The subsequent

rows report the corresponding optimal values of window sizes and amplitude thresholds. Values

shown are mean and standard deviation across all participants.

Right Left Right Left Right Left Right Left

Maximum

balanced accuracy0.94 ± 0.15 0.93 ± 0.14 0.61 ± 0.19 0.71 ± 0.21 0.93 ± 0.15 0.92 ± 0.14 0.89 ± 0.18 0.89 ± 0.18

Window size (ms) 414 ± 147 435 ± 138 408 ± 169 398 ± 141 394 ± 108 420 ± 110 405 ± 152 471 ± 167

Threshold (mV) 0.21 ± 0.11 0.23 ± 0.16 0.21 ± 0.13 0.16 ± 0.08 0.18 ± 0.11 0.17 ± 0.09 0.17 ± 0.11 0.20 ± 0.13

LateralGastrocnemiusTibialis Anterior VastusLateralis BicepsFemoris

Page 44: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

30

Figure 3-3 Balanced accuracies (BACC) averaged across all participants, showing degree of

EMG and MMG signal alignment as a function of amplitude threshold (vertical axis) and

window size (horizontal axis). BACC closer to 1 indicates greater coincident activity between

EMG and MMG.

During the gait cycle (Figure 3-4), coincident activity between EMG and MMG for the TA, VL,

and BF was observed primarily after heel strike (0-20%) and through swing phase (60-100%).

Despite these similar patterns, there was a significant interaction between modality (EMG and

MMG) and gait cycle division for the TA, LG, and VL muscles bilaterally, and the BF for the left

side only (p<0.001), which indicates differences between EMG and MMG at different subintervals

of the gait cycle. For the TA muscle, subsequent paired t-tests for each division of the gait cycle

revealed significantly higher MMG than EMG power mid-cycle (p<0.001) and significantly lower

MMG at the end of gait cycle (p<0.05). This trend was more prominent on the right versus left

side (Figure 3-4). For the LG muscles bilaterally, we observed significantly higher MMG than

EMG power following heel strike and during swing phase (60-100%) (p<0.01), and significantly

lower MMG power mid-cycle where we observed the majority of the EMG activity (p<0.001).

Although the VL EMG and MMG patterns appeared similar for the given participant in Figure 3-

4, the mean EMG and MMG signal powers across participants differed mid-cycle for the right

Page 45: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

31

side. However, for the left-sided VL, we observed significantly higher EMG activity at the

beginning of stance, and significantly higher MMG activity at the beginning of swing phase

(p<0.05) (Figure 3-5). A similar trend was seen between left-side VL and BF muscle; however,

the BF shows significantly higher MMG activity following heel strike (p<0.01). There was no

significant interaction between modality and gait cycle division, and there was no significant effect

of modality on power of the right-sided BF muscle only. Findings from the subgroup analyses by

gender, age and body size were consistent with those derived from the entire sample.

Figure 3-4 The typical activity patterns are shown for one participant (P28), showing the mean

(black) with standard deviation (red) for all muscles of the right leg. The gait cycle begins at heel

strike and swing phase typically begins at 60% of the gait cycle.

Page 46: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

32

Figure 3-5 Mean EMG (clear box) and MMG (shaded box) signal power across all participants

for the right (top row) and left (bottom row) legs. The asterisks (*) specify significant differences

between EMG and MMG power within an interval of the gait cycle division.

3.5 Discussion

3.5.1 Coincident MMG and EMG Activity

When comparing the processed MMG and EMG for the shank and thigh agonist-antagonist

muscles, we observed bursts of MMG activity that corresponded to EMG activity. This MMG-

EMG correspondence was quantitatively verified by the high balanced accuracies for the TA, VL,

and BF muscles, and corroborates the qualitative reports of previous dynamic studies [88, 96]. For

example, in studying the relationship between leg extensor activity and work load during cycle

ergometry, Shinohara et al. remarked that the shapes of the rectified MMG “roughly

approximated” those of the EMG signals but with “some” time delay [96]. Additionally, Shinohara

et al. noted the presence of MMG signal activity during the “non-contraction phase” of the muscle

where no corresponding EMG was visible, attributing this discrepancy to vibrations propagating

from the antagonist muscles [96]. This antagonist muscle “noise” may in part account for the lower

balanced accuracies in our study. In studying the relationship between EMG, MMG, and exerted

power in incremental cycle ergometry, Perry, et al. [88] reported 10-seconds of raw EMG and

MMG activity, where it appears that MMG activity generally aligns with EMG activity, with

limited MMG activity between contractions. Although they did not discuss this relationship, they

Page 47: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

33

did conclude that patterns of MMG amplitude may be more useful than EMG for examining power

output during continuous, dynamic tasks [88].

3.5.2 Discrepant MMG and EMG Activity

Interestingly, for the LG muscle, we found a large discrepancy between the EMG and MMG

signals. The accuracy of detecting active regions was only about 65% and the distribution of power

over the gait cycle was visibly different between modalities (Figure 3-4). The EMG power was

concentrated around the 50% mark of the gait cycle, i.e., when the shank is rolling over the foot

and the ankle is plantar flexing to maintain forward motion, which is the typical profile of EMG

activity during self-paced gait [1]. In our study, we observed a large peak in MMG power following

heel strike, then again around 60% of the gait cycle, and a third peak following toe off and before

heel strike (Figure 3-4). The observed MMG activity before and after heel strike has not been noted

in previous EMG/MMG studies [88, 96]. Further, the observed MMG-EMG temporal

misalignment is beyond the known electromechanical delay in voluntary, dynamic studies [53,

84]. The MMG signal for the LG muscle may in fact comprise vibrations from the passive

movement of surrounding non-muscular, soft tissues, and contraction of the antagonist muscles.

The latter is suggested by the peaks we observed in the MMG power of the LG at the end of swing

phase and after heel strike, corresponding to periods of electrical activity of the TA (Figure 3-5).

Previous research examining EMG and MMG during cycle ergometry concluded that some of the

MMG amplitude may be occurring from incomplete muscle relaxation and passive muscle fibre

movement [96], which may explain some of the discrepancies in our study between EMG and

MMG signal power during varying phases of the gait cycle. That agonist and antagonist co-

activation is manifested in MMG was also previously noted, albeit in an upper extremity study

following eccentric exercise, where the authors argued that such co-contraction provides joint

stability [97].

3.5.3 Distribution of MMG Signal Power Over Gait Cycle

MMG reflects the lateral oscillations of active and passive parts of the series elastic component of

the musculotendinous complex (MTC) and in particular, the muscle fibres that create mechanical

force [41-43]. Based on the force-velocity relationship, force is optimal at a certain muscle length

during isometric and slow concentric actions; however, that relationship is more complex in

dynamic actions [68]. The muscle-tendon complex is made up of the muscle fibres and the attached

Page 48: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

34

tendons. Muscle fibres transmit force to and interact with the tendons given tendon compliance.

Therefore, it is important to understand the relationship between the lengths of both muscle fibre

and tendon during dynamic activities. In their measurements of fascicle length change and EMG

activity of the TA and VL during walking, Chleboun, et al. [68] found significant fascicle length

change and bursts of EMG activity in the TA between 60-80% of the gait cycle. Similarly, for the

TA muscle, we found an aggregation of MMG signal power during the first 20% of the gait cycle

and again around 60%. Furthermore, in the VL, Chleboun, et al. [68] observed significant length

change and EMG activity at the end of the gait cycle (75-100%); however, fascicle length did not

exhibit significant length change in the first or second portions of the gait cycle. Similarly, in our

study, we observed an accumulation of MMG signal power for the VL muscle in the initial and

final 20% of the gait cycle (Figures 3-4 and 3-5). Thus, our observations of MMG signal power

distribution over the gait cycle, can in part be understood in terms of underlying fascicle length

changes.

3.5.4 Differences Between MMG and EMG Signal Power Distribution Over the Gait Cycle

We found lower coincident EMG and MMG activation for the LG, and significant differences

between the modalities in the various stages of the gait cycle (Figure 3-5). Prior to the swing phase,

we observed substantial electrical activity as the gastrocnemius activated to assist with push-off.

In their study using ultrasonography to examine MTC during various dynamic movements,

Fukunaga, et al. [77] showed that during the stance phase of walking, the medial gastrocnemius

(MG) muscle fibres maintained constant length while the tendon stretched, whereas during push-

off, both MTC and tendon shortened rapidly. In our study, we saw high EMG power at the end of

stance phase (40-60%) with low corresponding MMG power (Figure 3-5). At this point of the gait

cycle, the gastrocnemius muscle is contracting nearly isometrically [77], where we typically see

lower peak-to-peak MMG amplitudes in comparison to concentric and eccentric components of

contraction [43, 63]. Thus, we should expect lower MMG activity during this period of electrical

activity. Furthermore, the Achilles tendon lengthens during stance and recoils in push-off [78],

potentially contributing to the MMG signal at these times. Additionally, we saw differences in

modality-specific power between left and right legs for the TA, VL and BF muscles, which may

be related to participants walking with the left leg always on the inside of the oval track. Although

we know that EMG muscle activity must be adjusted in order to make a turn during gait, overall

Page 49: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

35

gait patterns remain unchanged [98]. Nevertheless, these findings suggest that the relationship

between EMG, MMG and the associated temporal delay may be more complex during dynamic

movements and future studies ought to consider muscle-specific MTC length changes in

interpreting lower limb MMG during gait.

3.5.5 Limitations and Future Work

MMG signal characteristics may vary between adjacent muscles, specifically medial versus lateral

gastrocnemius. Although some studies show comparable EMG activity between the LG and MG

heads when recording different speeds of walking and running [1, 99, 100], other studies focus

only on MG given its larger muscle belly and greater activation during self-paced walking [101].

Furthermore, differences in LG and MG muscle fibre architecture, such as muscle fascicle length,

their insertion angle, and muscle thickness, suggests that there may be differences in the MMG

signals of these two muscles [102, 103]. We did not see any effects of gender, age, or body size

on MMG power distribution and timing. However, future research should investigate potential

effects of age and body size on MMG frequency responses in dynamic contractions [86] in

developing and advanced aging populations. Future studies should also investigate MMG

differences between the MG and LG muscles as manifested during gait.

Despite the aforementioned inter-contraction noise, the average MMG signal for the TA muscle

revealed two distinct peaks corresponding to one EMG burst, one at the start and the other at the

end of the contraction (Figure 3-4). Beck, et al. [59] observed, via wavelet transform, MMG

spectral differences between the concentric and eccentric components of isokinetic contractions.

Thus, our finding of dual MMG peaks, may suggest that spectral filtering of MMG has potential

to differentiate between concentric and eccentric components of the contraction in gait. This would

be particularly important given that EMG spectral analysis does not distinguish between concentric

and eccentric contractions [63, 104, 105]. Wavelet-based MMG analysis for automatic detection

of concentric and eccentric contractions during gait thus merits further exploration.

3.6 Conclusions

During pediatric gait, electro- and mechano-myographic activation patterns appear to be

temporally aligned for the tibialis anterior, vastus lateralis and biceps femoris muscles of both legs.

Page 50: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

36

However, for the lateral gastrocnemius muscle, the EMG and MMG activations exhibit large

temporal discrepancies, well beyond that attributable to electromechanical delay. Passive

vibrations of nearby tissues and vibrations of the antagonist muscles may contribute to the

observed EMG and MMG activation offsets at the LG muscle. Differences between EMG and

MMG signal power distributions over the gait cycle may be related to fascicle

elongation/shortening and muscle-specific musculotendinous complex length changes during gait.

The reported discrepancies between EMG and MMG temporal distributions of signal power

suggest a complementary role for MMG in identifying and tracking electro-mechanical changes

in musculotendinous behaviour due, for example, to injury, disease or training.

3.7 Acknowledgements

The authors would like to thank NSERC Create CARE program for funding the primary author,

and Ka Lun Tam and Pierre Duez from the Prism Lab for their assistance with instrumentation and

coding.

Page 51: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

37

Chapter 4

Mechanical Synergies During Gait as Revealed Through Mechanomyography

4.1 Abstract

This paper investigates mechanical synergies during gait as revealed through lower limb

mechanomyography (MMG). We recruited 10 typically developed adults and recorded

simultaneous electromyography (EMG) and MMG of the tibialis anterior, medial and lateral

gastrocnemius, and vastus lateralis muscles during treadmill walking and running, each at 2

different speeds. Synergies were extracted from EMG and MMG signals during the gait cycle

using non-negative matrix factorization for each condition. On average, 2.49 ± 0.53 (VAF 96.60

± 0.81) synergies were extracted from EMG signals across all conditions and participants,

consistent with previous research. In contrast, only 1.70 ± 0.64 (VAF 95.95 ± 0.64) mechanical

synergies were extracted from the corresponding MMG signals across all conditions. Interestingly,

all extracted mechanical synergies captured muscle co-activation; however, there appear to be

distinct activation trends between walking and running conditions.

Page 52: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

38

4.2 Introduction

Gait is a complex task involving both voluntary and automatic processes within the nervous and

musculoskeletal systems. To efficiently organize movements and actions, it is suggested that the

central nervous system (CNS) deploys simplifying neural commands called muscle synergies, to

efficiently control multiple muscles, rather than send separate commands to individual muscles

[4]. Muscle synergies provide a mechanism by which task-level motor intentions are translated

into detailed, low-level muscle activation patterns [106]. A small number of muscle synergies may

be invoked in varying combinations to produce a wide variety of motor behaviors [107].

When examining electromyography (EMG) data in gait, Ivanenko, et al. [100] showed that from a

sample of 25 muscles, gait could be explained by a combination of five basic synergies. Similarly,

other studies have shown that muscle synergies are motor task-specific, and variations in

movement, such as in reacting to a perturbation [106] or transitioning between walking and

running [108], are modulated by activation levels rather than activation patterns. However, muscle

synergies measured via EMG are limited because they only reveal the electrical, or neural aspect

of muscle function and do not convey the mechanical behavior of the muscles [43, 84, 109, 110],

which when paired with the electrical information can provide insight into sensory-motor

coordination [111]. Indeed, one can interpret the neural signal (EMG) as the input to the muscle

and the mechanical behaviour (MMG) as the output of the contraction [92]. A complementary

synergy analysis of the mechanical activity of muscles may thus be informative in studying

neuromotor control of movement [112] or its alteration with aging [113], the effect of different

exercise interventions on neuromuscular responses and force production [114], or neuromuscular

changes accompanying neurological disease and its mechanical consequences [92].

Mechanomyography (MMG) reflects the lateral gross movement of muscle fibres along with the

subsequent vibrations at the muscle’s resonant frequencies and the dimensional changes of active

muscle fibres [22, 24]. MMG is said to be the mechanical counterpart to EMG, and has been used

to describe motor control strategies, in terms of motor unit summation [10], firing pattern during

fatigue [11], and force generation during contractions [12]. Although MMG has been used as a

complementary modality to EMG in isometric and simple isokinetic contractions [32, 43], the

MMG response reflects both active and passive properties of muscle [115], which are in turn

affected by joint movement during dynamic motor tasks [22, 116]. Furthermore, our previous work

Page 53: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

39

has shown that the relationship between EMG and MMG is more complex during dynamic

movements, giving rise to modality-specific activation patterns during the gait cycle [13]. Ting

and McKay [106] recommended that because of the interaction of musculoskeletal elements during

movement, neural commands and muscle actions should not be studied in isolation, but rather as

part of a neuromechanical model. To this end, it is important to consider the coordination of

mechanical muscle function.

Several studies have reported similar muscle synergies between walking and running as

determined through EMG [117-119]. Ivanenko, et al. [100] and Hagio, et al. [108], reported that

the neural control of both walking and running share 4 or 5 common synergies. These synergies

flexibly generate different movement patterns by modulating the weighting factors and synergy

coefficients associated with each synergy. To the best of our knowledge, there have been no

studies evaluating MMG-based muscle synergies. We thus set out to determine mechanical

synergies as manifested in mechanomyography during treadmill walking and running. Based on

mechanomyographic evidence of contemporaneous agonist and antagonist co-contraction [97,

116] and previously identified differences between EMG and MMG patterns during the gait cycle

[55], we expected that MMG-based synergies during gait may differ from those reported in EMG

research.

4.3 Methodology

4.3.1 Participants

We recruited 10 healthy adults (3 males and 7 females, 28 ± 9 years old, height: 165.5 ± 6.1 cm,

weight: 64.8 ± 12.9 kg). Participants reported no known neurological disorders, or musculoskeletal

injuries (current or previous) limiting their walking. Participants wore athletic shoes and shorts

during the study. The research ethics boards of Holland Bloorview Kids Rehabilitation Hospital

and University of Toronto approved the study protocol, and all participants provided informed

written consent.

Page 54: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

40

4.3.2 Data Collection and Instrumentation

Wireless surface electromyography (EMG) sensors (Trigno by Delsys Inc, Boston, MA, USA)

were fastened with double-sided tape over the muscle belly of each of the following lower limb

muscles: tibialis anterior (TA), lateral gastrocnemius (LG), medial gastrocnemius (MG), and

vastus lateralis (VL). MMG data were collected using tri-axial accelerometers (ADXL337, Analog

Devices Inc, Norwood, MA). The accelerometers were positioned on the muscle bellies, usually 3

cm proximal to the EMG sensor, and attached with medical grade tape. For each muscle, the

accelerometer was oriented such that the z-axis was perpendicular to the longitudinal axis of the

muscle.

An ultra-thin, force-sensitive resistor (FSR) (FSR 406, Interlink Electronics, Shenzhen, China)

was inserted inside the participant’s shoe to record heel strike for each step [14]. All data were

collected bilaterally. Each sensor was connected to a data logging system housed inside a backpack

(total mass: 1.83 kg) worn by the participant. All analog data were acquired via a custom-made

LabVIEW program at a sampling rate of 1 kHz for MMG and FSR, and 2 kHz for EMG.

4.3.3 Experimental Setup

Prior to the walking trials, treadmill safety was reviewed with participants. A 30 second baseline

of quiet stance was collected. Participants were then given a 2-minute period to warm-up on the

treadmill (GK200T, GaitKeeper Rehab Treadmills, 2014 Mobility Research) at a self-selected

speed. Participants performed four 1-minute treadmill trials at the following speeds: 3, 4, 5, and 6

mph. These speed trials will be referred to as: 3 mph = slow walk (SW), 4 mph = fast walk (FW),

5 mph = slow run (SR), and 6 mph = fast run (FR), respectively. Participants rested for 30-60

seconds between trials as preferred, and completed a cool down period of a few minutes following

the last trial. Participants did not complain of fatigue or discomfort from wearing the sensors and

backpack, and all participants were able to complete all treadmill trials.

4.3.4 Data Pre-Processing

To extract MMG content from the accelerometry signals, the z-component of the signal was

bandpass filtered (5th order Butterworth filter) between 5-50 Hz and full-wave rectified [13]. EMG

data were bandpass filtered (4th order Butterworth filter) between 30-500 Hz, full-wave rectified,

Page 55: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

41

and down-sampled to 1 kHz [15]. Both MMG and EMG signals were smoothed using a moving

window average of 101 samples. For each trial, heel strike was identified based on the FSR sensors

and used to demarcate the gait cycle. Each gait cycle was down sampled to 200 points.

For each condition, the signals from the first and last 10 gait cycles were removed to eliminate

initiation and termination effects. Subsequently, 15 strides from each leg were used to build a

matrix for synergy extraction; each matrix consisted of 8 rows (four muscles each for right and left

legs) x 3000 columns (15 gait cycles x 200 data points). The data in each matrix were amplitude

normalized by the maximum value across all 4 matrices (one for each condition) for each subject

so that all values ranged from 0 to 1. The data were further normalized by the within-matrix

standard deviation to have unit variance, thus ensuring that all muscles were equally weighted

[120]. All data analysis was carried out via custom-designed scripts and the NMF toolbox for

MATLAB [16].

4.3.5 Muscle Synergy Extraction

Synergies were extracted from seven different groupings of the data from the four conditions: Slow

Walk (3 mph), Fast Walk (4 mph), Walk (3 + 4 mph), Slow Run (5 mph), Fast Run (6 mph), Run

(5 + 6 mph), Global Walk + Run. For Walk and Run groups, the NMF matrix was created by

combining the slow walk and fast matrices together (8 x 6000 matrix), and the global walk + run

combined all four conditions (8 x 12000 matrix).

A non-negative matrix factorization (NMF) algorithm was used to extract muscle synergies for

each condition [121]. The technique assumes that a muscle activation pattern (M) is comprised of

a linear combination of a few muscle synergies recruited by a time- varying coefficient (C). The

recruitment coefficient represents the neural command that specifies how that synergy is

modulated over time, and how much each synergy contributes to a muscle’s total activity pattern

[4, 16-18]. Let W represent the muscle synergy matrix, with each column representing a muscle

synergy, and C be the synergy activation coefficients. Muscle activation pattern, M, can then be

expressed as,

Page 56: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

42

𝑀𝑚×𝑛 = 𝑊𝑚×𝑠𝐶𝑠×𝑛 (1)

where m is the number of muscles (m = 8), n is the length of the gait cycle pattern (e.g., n = 3000

in the Slow Walk condition), and s is the number of muscle synergies. M, W and C are represented

as:

𝑀 = [𝑀1(𝑡)

⋮𝑀𝑚(𝑡)

], 𝑊 = [ 𝑊11 ⋯ 𝑊𝑠1

⋮ ⋱ ⋮𝑊1𝑚 ⋯ 𝑊𝑠𝑚

], 𝐶 = [𝐶1(𝑡)

⋮𝐶𝑠(𝑡)

] (2)

Using the calculated coefficients and weighting factors, a reconstructed muscle activation pattern

is given by:

𝑀𝑟 = 𝑊∗𝐶∗ (3)

where W* and C* represent synergy weighting factors and synergy activation coefficients.

Typically, synergies are retained such that and comprise a compact representation

of the original data.

The NMF was implemented using an iterative optimization. The matrices W and C, are randomly

initialized, and are iteratively updated such that the squared error between the original and

reconstructed data is minimized, i.e., . The goodness-of-fit between

reconstructed and original muscle signals was measured using the correlation of determination (r2)

and the muscle-specific and overall variability accounted for (VAF) at each condition. The number

of extracted synergies required to meet the threshold VAF will be termed the synergy level, .

4.3.6 Walking vs. Running

To examine the concordance between weighting factors among conditions, especially grouped

walking and running, we used a cosine similarity analysis [122]. When comparing two muscle

synergies, the inner product of the two muscle synergy vectors (divided by the product of their

norms) was compared between two conditions, representing the cosine of the angle between the

vectors. As such, an inner product closer to one represents greater similarity between the vectors,

thus similarity between synergies. A critical threshold of cosine similarity (r > 0.7682 ± 0.01)

s *W*C

2

,W Cmin M WC

Page 57: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

43

was determined based on a Monte Carlo simulation of uniformly distributed random data with two

synergy levels that was performed on n = 1000 data sets [120, 123].

4.4 Results

All of the participants were able to perform both walking and running conditions, and all

participants spontaneously shifted their gait pattern from walking to running at 5 mph.

4.4.1 Electro-mechanical muscle activity in gait

EMG and MMG patterns for a representative participant can be seen in Figure 4-1. Overall during

walking conditions, we observed bursts of TA and VL electrical activity in the first 25% of the

gait cycle, and then activity ramped back up at the end of the gait cycle. A small peak of TA

activity was seen around 75% of the gait cycle. The MG and LG exhibited activity between 30 and

60% of the gait cycle. When examining the mechanical aspect, MMG activity for the TA and VL

muscles followed similar trends as EMG; bursts of activity occurred between initiation and the

first 25% of the gait cycle, followed by a small burst of activity around 50-70%, and an increase

in activity around 85%. The LG and MG exhibited similar MMG activity as the TA and VL at the

start and end of the gait cycle; however, more mechanical activity was observed between 55% and

75% of the gait cycle, especially of the LG.

During running, we observed that most of the EMG muscle activity shifted to the first 40% of the

gait cycle, except for the TA. EMG of the TA had two peaks, one around heel strike and a second

peak between 40 and 75% of the gait cycle. In terms of MMG, a similar trend of activity around

heel strike was observed; however, there were larger bursts of activity beginning at around 75%,

peaking around heel strike, then ending around 25% of the gait cycle. Similar to EMG patterns,

the second MMG peak tended to shift towards the first half of the gait cycle, and was associated

with a decrease in amplitude.

These results are in line with previously reported EMG data during walking and running [108, 120,

123], and similarly, MMG findings are comparable to previously recorded MMG during self-paced

gait in typically developing children [76, 99, 124].

Page 58: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

44

Figure 4-1 An example of EMG and MMG during one gait cycle from the right leg recorded

during each of the treadmill speeds (slow walk (SW), fast walk (FW), slow run (SR), and fast run

(FR)) for a representative participant (P10).

4.4.2 Extracting Muscle Synergies

The reconstruction of neural and mechanical muscle signals via NMF analysis was assessed in

terms of VAF value [55]. The scree plot for the overall VAF is shown in Figure 4-2. For EMG,

the lowest VAF across all groups was 74%, whereas the lowest VAF for mechanical synergies was

91%. The VAF of the reconstructed EMG signals continued to increase with the inclusion of more

than two or three synergies, whereas with MMG signals, >90% variance is accounted for with only

Page 59: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

45

one or two synergies. We extracted the lowest number of synergies based on an overall VAF cutoff

of 95% for both neural and mechanical data, as well as a minimum muscle VAF cutoff of 80% to

ensure that all muscles were well reconstructed in the overall synergy [121, 122].

Our EMG synergies show that across all conditions and participants, the average number of

extracted neural synergies was 2.49 ±0.81 at a mean VAF of 96.0 ±0.81%. We saw higher number

of synergies with walking (SW, FW) than running (SR, FR) conditions. In fact, on average, we

extracted 2.55 ±0.52 (VAF 96.46 ±0.72%) muscle synergies for walking (W) conditions, 2.27

±0.47 (VAF 96.32 ±0.79%) muscle synergies for running (R) conditions, and 2.73 ±0.47 (VAF

96.81 ±0.96%) for grouped walking and running (W+R) conditions, respectively. Across all

participants and groups, two or three synergies were extracted, and there was only one participant

in the FR condition where four synergies were extracted.

The average number of extracted mechanical synergies was 1.70 ± 0.64 at a mean VAF of 95.95

± 0.64% across all conditions and participants. Across grouped conditions, we extracted an average

of 1.55 ± 0.52 (VAF 95.99 ± 0.71) muscle synergies for walking (W), 1.91 ± 0.52 (VAF 95.97 ±

0.44) muscle synergies for running (R), and 1.82 ± 0.60 (VAF 95.88 ± 0.49) for grouped walking

and running (W+R). In most cases, one or two synergies were extracted across participants;

however, in seven cases, three mechanical synergies were extracted: one in SR, three in FR, two

in R and one in the global condition. For one participant, only one synergy was extracted across

all conditions – this synergy showed nearly full activity of all the muscles at once.

Page 60: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

46

Figure 4-2 Scree plot of overall VAF (across participants) for each synergy level. Each row of

plots corresponds to VAF values for one condition: slow walk (SW), fast walk (FW), slow run

(SR), fast run (FR), walk (W), run (R), and global walk + run (W+R).

4.4.3 Neural Synergies

The extracted neural synergy weights and synergy coefficients for the grouped conditions are

shown for a representative participant in Figure 4-3. Across all conditions, we saw a consistent

pattern of medial and lateral gastrocnemius activity in Synergy 1, tibialis anterior activity in

Synergy 2, and vastus lateralis activity in Synergy 3 (Global W+R in Figure 4-3). For some

Page 61: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

47

participants, synergy levels were switched or combined in certain conditions, a phenomenon

previously reported in the literature [117, 125]. For example, in the Run condition in Figure 4-3,

where Synergy 1 is made up of the TA-dominant synergy instead of the LG/MG synergy dominant

in the Global and Walk conditions. Similarly, the second and third synergies comprised co-activity

of LG/MG and VL activity; however, VL is still predominant in Synergy 2 and LG/MG are

predominant in Synergy 3. In cases where only two synergy levels were retained, the first synergy

was commonly composed of gastrocnemius and vastus activity, while the second synergy

consisted primarily of TA activity.

Figure 4-3 An example of extracted neural synergies for the grouped conditions for one

representative participant (P08). At each synergy level, we show the corresponding synergy

weights (W) at each muscle (left (blue) and right (red) sides) and the mean synergy coefficients

(C) that together account for at least 95% of the reconstructed muscle signals.

4.4.4 Mechanical Synergies

Mechanical synergies were extracted based on MMG activity and a minimum overall VAF set at

95%. At this threshold, between one and three synergies were extracted across all walking and

running conditions. When we look at the average VAF across all participants and conditions, two

synergies should be extracted for all conditions, which is represented for a typical participant

across all conditions in Figure 4-4. On average, Synergy 1 predominantly consisted of TA and VL

activity, whereas Synergy 2 involved more LG and MG activity. This trend was consistent across

Page 62: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

48

all conditions for this participant, except grouped Walk, where we found more LG activity in

Synergy 2. Interestingly, there is a large burst of LG activity, while MG activity is only around

50% active, and the VL is active at around 80%. This is uncommon – typically, the LG and MG

are grouped together in one synergy, whereas the TA and VL are grouped together in the second

synergy. Similarly, the gastrocnemius muscles are commonly grouped together as they move

synergistically.

The Global synergies were extracted from all four grouped conditions (i.e., SW, FW, SR, FR), and

there was a dominance of TA and VL in the first synergy (Figure 4-4). Interestingly, in neural

synergies, we saw a dominance of LG and MG in the first synergy, and TA and VL in the second

synergy, which was reversed in MMG. When looking at the mechanical Global synergies, the TA

and VL synergists are more active in Synergy 1 and the LG and MG synergists are active in

Synergy 2. Interestingly, the Global synergy is more similar to the Run condition, suggesting that

there is a larger influence of Running synergies in the Global representation of mechanical activity.

When looking at the synergy coefficients across all conditions, there was a peak of activity

following heel strike (gait cycle 0%) and around 60% of the gait cycle in Synergy 1. The temporal

activity at this synergy is consistent across all conditions. In Synergy 2, there was a small peak

following heel strike, then a second peak near the end of the gait cycle. In walking conditions, the

second peak was closer to the end of the gait cycle (>90%), whereas in running, this peak of activity

spiked earlier in the gait cycle (around 75%). This is consistent with EMG that is activated earlier

in the gait cycle during running compared to walking [99]. In the global condition, the peak is

evident around 75% of the gait cycle, however the amplitude is visibly attenuated in comparison

to the individual walking and running trials. These differences may be indicative of kinematic

differences observed between walking and running.

Page 63: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

49

Figure 4-4 Mechanical synergies extracted for all conditions for a representative participant (P08).

Muscles are grouped together with left (blue) and right (red) bars.

4.4.5 Walk vs. Run

Unlike EMG muscle synergies for walk and run conditions, extracted MMG synergies differed

between walking and running. In walking, we saw high LG (0.8) and MG (1.0) activity with only

about 0.25 activation of TA and VL each in the first synergy, and about 0.75 activation of TA and

VL, with about 0.3 activation of LG, in the second synergy. In contrast, in running, we see

predominantly TA, LG and MG activity in the first synergy, and high activity of LG in the second

synergy. Furthermore, the walk + run group did not appear as a combination of both synergy levels,

but rather as a more balanced pattern of predominantly high TA and VL activity in the first synergy

and high LG and MG in the second synergy.

Page 64: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

50

When comparing Walking and Running conditions, EMG synergies show a distinct difference

within synergy levels in each condition (as depicted by blue areas in Figure 4-5; cosine similarity

lower than the critical threshold of r = 0.7682 ± 0.01) and high similarity between synergies

extracted from Walking and Running conditions (as depicted by red areas in Figure 4-5). In

contrast, the cosine similarity for MMG synergies showed high degrees of similarity across and

between synergy levels (cosine similarity greater than the critical threshold of r = 0.7682 ± 0.01).

Even though there is a great deal of similarity between and within conditions, there was also a

large amount of variability between participants and trials, as is evident with the lack of unifying

pattern in Figure 4-5.

Figure 4-5 Cosine similarity matrix of EMG (left) and MMG (right) synergy weights for grouped

walk vs. run conditions. EMG synergies show distinct patterns between synergy levels (syn1, syn2,

syn3), whereas MMG synergies show a lot of similarity between synergies and conditions (more

red areas).

4.4.6 Reconstruction of Muscle Signals

The original and reconstructed EMG and MMG muscle signals based on the Global W+R

condition for one representative participant are seen in Figures 4-6 and 4-7. On average, EMG

signals were reconstructed to about 85.07 ±9.78% VAF of the original signals; whereas MMG

Page 65: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

51

signals were only reconstructed to about 81.38 ±8.11% VAF of the original signals across all

conditions and participants. A Shapiro-Wilk test for normality indicated that goodness-of-fit

showed a non-Gaussian distribution across conditions. Friedman’s non-parametric rank test with

Wilcoxon Signed-Rank post-hoc test for multiple comparisons (α = 0.05) was then applied for

block analysis of variance to determine if there was an effect of condition on goodness-of-fit for

mechanical and neural synergies across participants. All statistical analyses were performed using

RStudio for R (Version 1.0.136, RStudio, Inc.). For EMG signals, there was no effect of condition

on goodness-of-fit between reconstructed and original muscle signals (p=0.79). However, there

was a significant effect of condition on reconstruction error of MMG muscle signals (p<0.05).

Post-hoc tests showed significant differences in goodness-of-fit between Fast Walk and Fast Run

conditions (p=0.016) and Fast Run and Grouped Walk conditions (p=0.042).

Figure 4-6 - In the NMF analysis, each original muscle signal (dotted line) is reconstructed (black

line) based on the synergy weights and synergy coefficients (coloured lines) through the gait cycle.

Shown are the reconstructions for EMG for Global W+R for P01.

Page 66: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

52

Figure 4-7 - Reconstructed MMG signals based on two synergy levels in the NMF analysis.

Shown is P01 based on the Global condition.

4.5 Discussion

This study presents an analysis of the spatiotemporal coordination of mechanical activity during

gait using non-negative matrix factorization decomposition. Understanding mechanical patterns

during motor tasks is useful in detecting pathology, providing feedback about movement patterns,

and in controlling assistive technologies and robots.

4.5.1 Electromechanical Activity during Gait

The spatiotemporal patterns of both mechanical and electrical aspects of muscle contraction are

comparable to previously reported data [55, 99, 109, 124]. Previously, in self-paced gait in youth,

we reported a discrepancy between gastrocnemius mechanical and electrical activity [55], which

is again observed in the present study. We proposed that this discrepancy between

electromechanical activities might be related to muscle architecture or gait maturation; however,

Page 67: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

53

the current results maintain this observed discrepancy in both heads of the gastrocnemius in adult

participants. MMG demonstrates the length change of muscle fibres and the resulting vibrations

of the surrounding soft tissues during contraction [22, 23]. Additionally, studies have shown that

different length changes occur across different movements, and as such, the MMG reflects both

the active and passive length change in the contractile components [116]. Our results align with

fascicle length changes seen in gait in previous studies [68, 77, 78]. This reinforces our findings

that the MMG data reflect length changes of the muscle fibres during dynamic movements.

As gait transitions from walking to running, increases in speed are associated with an increase in

mechanical activity due to observed changes in kinematics, kinetics and muscle activity [126, 127].

During the gait cycle, we saw that EMG activity was dominant in the first half of the gait cycle

(Figure 4-1). However, we did not see this activity reflected in the measured MMG during running.

In fact, there was a decrease in MMG activity during the middle of the gait cycle (Figure 4-1).

When studying the muscle fascicle length changes of the triceps surae group, although the trends

were similar, there were differences between walking and running, suggesting fewer fascicle

length changes and more length changes in the tendinous complex during running [79, 128].

Additionally, as running speeds increase, we see less joint movement and less sarcomere length

changes [76], as well as a greater dependence on the elastic properties of the tendons to increase

movement efficiency [75], which may explain the decrease in MMG activity observed in our study.

As running speed increases, there is a greater dependence on the stretch-shortening cycle to

improve biomechanical efficiency [126].

4.5.2 Muscle Synergy Analysis

Most commonly, an NMF algorithm has been applied to decompose muscle activity and other

various large-scale data in neuroscience, computation biology, and image and audio processing

[129]. In this analysis, a high-dimensional data matrix is linearly decomposed into a low-

dimensional basis vectors and scaling coefficients that are more easily interpretable [129]. When

applied to surface EMG, the extracted muscle synergies provide a way of understanding how the

CNS organizes and recruits groups of muscles during various movements [107]. Previous research

has shown that walking and running can both be explained by common neural commands as

reflected in consistent muscle synergy patterns found in both children and adults [108, 125].

Page 68: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

54

The neural synergies extracted in this study show results in line with previous work showing that

a few common synergies can be used to explain both walking and running gait patterns. Synergies

extracted from individual conditions were consistent with those synergies extracted from grouped

running or walking, as well as a global condition encompassing all gait patterns. This demonstrates

that the coordination of electrical activity measured through surface EMG is consistent between

movement patterns. In contrast to some other studies that extract four or five synergies [99, 100],

we extracted only two or three neural synergies. Since NMF analysis reflects the structure of the

data rather than the experimental design, the robustness of our synergies is limited by the input of

only four muscles. Some studies suggested that including more muscles in a synergy analysis

extraction might significantly affect the structure and number of muscle synergies [4, 20]. Despite

this, our results show common patterns in neural commands that direct the complex patterns of

muscle activity in dynamic movements. Although EMG is the common modality used to study

muscle activity, the limitation with this modality is that we cannot directly infer mechanical

properties of the muscles themselves.

Neuromechanical theories of motor control stress the importance of studying the neural input

alongside the mechanical output in order to better understand control of the whole movement. As

such, in this study, we proposed a synergy analysis that assesses the control and coordination of

mechanical activity as measured by accelerometer-based MMG. In comparison to neural

synergies, we extracted fewer dimensions of mechanical synergies, extracting only one or two

levels (>95% overall VAF) for most gait conditions and only three synergies in some cases (n=7).

When extracting two mechanical synergies, the weighting factors showed large amounts of co-

activity among muscles in each synergy (Figure 4-4), resulting in high degrees of similarity

between synergy levels (Figure 4-5). This is a departure from the pattern of neural synergies, which

pinpointed distinct groups of muscles at each synergy level, with minimal co-activation of other

muscles (Figure 4-5). This co-activity between synergies may be related to the concurrent

lengthening and shortening of muscle fibres during active and passive joint movement [130].

Several groups have studied the behavior of human muscles with live ultrasound during walking

and have observed the lengthening of muscle fascicles and the presence of EMG activity [68, 75-

78]. These studies showed that during gait, EMG activity may not be directly accompanied by

lengthening of the muscle fascicles – in fact, the gastrocnemius contracts isometrically and

lengthens in absence of electrical activity [77, 78], and similarly the tibialis anterior and vastus

Page 69: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

55

lateralis show some periods of near isometric contraction [68]. If we consider that MMG measured

in complex, dynamic motor tasks to be related to the lengthening and shortening of muscle

fascicles, our MMG spatiotemporal patterns show peaks of activity that align with muscle fascicle

length changes during the gait cycle. Specifically, our recorded MMG data for walking and

running show a large peak of activity increasing before heel strike across all muscles, followed by

an MMG amplitude decrease subsequent to heel strike (Figure 4-1). This is in line with previous

work by Lichtwark, et al. [79], who showed that during walking, the gastrocnemius muscle fibres

maximally lengthen at heel strike, then shorten rapidly until the foot is flat on the ground surface.

At the same time, the ankle plantar flexes and the whole musculotendinous complex also shortens,

which may contribute to an increase in MMG amplitude. Another group studying the relationship

of MMG and joint angle during maximum voluntary contraction hypothesized that changes in

MMG amplitude may be related to the joint angle differences in mechanical properties of

contraction and/or slack in the muscle [131]. As such, MMG may reflect the combination of

muscle fascicle behavior and joint angle changes during gait, and since these movements are

occurring simultaneously at certain parts of the gait cycle, we can expect the coordination of

mechanical patterns between muscles to be similar. This is reflected in our results, where we see

higher goodness-of-fit between reconstructed and original signals at lower synergy levels in

comparison to neural signals.

Our findings suggest that a lower dimension of control than neural activity governs mechanical

muscle activity. This may in turn be related to the need for creating efficient movement patterns.

We know that the human body is designed to optimize the efficiency of movement. For example,

muscle and tendon interactions allow us to exploit the elastic properties and the stretch-shortening

cycle to increase the force generating capacity of muscle that is utilized in running [77]. Perhaps

the observed alignment of MMG responses across muscles during walking and running are

indicative of efficient movements where the muscle fascicles and joint angles move in unison.

Consequently, efficient movements should be represented by very low dimensional mechanical

synergies. Alternatively, movements that are inefficient, where muscle fascicle length changes

follow more atypical patterns, such as with spastic gait, mechanical synergies would be more

complex. Therefore, we may expect inefficient, pathological movement patterns to generate more

synergy levels than efficient, typical movement patterns. In this way, mechanical synergies could

be applied in clinical diagnostics, rehabilitation tracking, and in measuring athletic performance.

Page 70: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

56

4.6 Conclusions

This is the first study to analyze muscle synergy during walking and running incorporating both

neural and mechanical aspects of muscle function via EMG and MMG, respectively. Our findings

contribute to a fuller understanding of the neuromechanical control of selected lower limb muscles

during treadmill gait. EMG and MMG should be used as complementary modalities in research

since, together, they provide an understanding of the neural control and mechanical output of

muscle contraction, coordination, and movement. Future work should focus on investigating

mechanical synergies in populations with gait disturbances, and to expand the analysis to other

movements, such as reaching tasks.

4.7 Acknowledgments

The authors would like to thank NSERC Create CARE program for funding the primary author,

and Ka Lun Tam and Pierre Duez from the Prism Lab for their assistance with instrumentation and

coding.

Page 71: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

57

Chapter 5

Designing a Wearable MMG-based Mobile App for Gait Rehab

The entirety of this chapter is reproduced from the following manuscript: K. Plewa, M. Silverman,

S. Orlandi, T. Chau, M.H. Thaut, Designing a wearable MMG-based mobile app for gait rehab,

in: IEEE Life Sciences, IEEE Xplore, Sydney, Australia, 2017.

This is an author-created, un-copyedited version of an article published in in IEEE Xplore. IEEE

is not responsible for any errors or omissions in this version of the manuscript or any version

derived from it.

© 2017 IEEE DOI 10.1109/LSC.2017.8268187

5.1 Abstract

Movement disorder therapies involving sonography and rhythmic entrainment have shown lasting

improvements to gait dynamics. Although optimal parameters for gait training have yet to be

defined, past studies have shown that increasing training frequency enhances neural

reorganization, thus supporting the development of wearable technologies in gait rehabilitation.

This paper presents a novel tool for the acquisition of muscle activity, their analysis, and

presentation as a live biofeedback signal that distinguishes between typical and atypical gait

patterns. Muscle activity is recorded and analyzed on an Arduino, then sent to an Android for

feature detection via Bluetooth. Auditory feedback will be presented as a fixed tempo based on

stride rate and an interactive drum kit based on matching gait patterns. By developing a tool that

can be used at-home, users will be able to train daily and maintain longer rehabilitation programs,

thus encouraging neural reorganization. This mobile app will allow us to improve quality of life

by enhancing training outcomes and functional gait dynamics.

Page 72: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

58

5.2 Introduction

Neurological rehabilitation makes use of various treatment modalities, including techniques from

neurologic music therapy, which has shown very positive impacts on learning and motor abilities

of children with neurological deficits [12]. The link between auditory and motor systems is evident

in the research, and its influences can be seen clearly when applying rhythmic entrainment to

movement disorder rehabilitation [14]. Studies providing fixed rhythmic auditory stimulus (RAS)

have shown improvements in gait patterns and stride parameters for patients with stroke,

Parkinson’s disorder, traumatic brain injury, and cerebral palsy [15-18]. In children with cerebral

palsy (CP), damage to the motor cortex disrupts normal processes for motor control thereby

affecting rhythmic movements. Studies have shown improvements to symmetry and stride rate

with both therapy-guided and self-guided RAS gait therapies in children with CP, suggesting the

need for at-home therapies [15]. Therefore, designing a gait intervention that is both wearable and

mobile accessible, improves access to rehabilitation and the potential for increased quality of life.

The study of muscle activity is important for a variety of clinical and rehabilitation applications,

including pathology diagnostics, prosthetic control, access technologies, and movement

biofeedback. Muscle activity can be measured using mechanomyography (MMG), a

complementary modality to electromyography (EMG), which describes the mechanical aspects of

muscle function based on the gross lateral movements of muscle fibers and the subsequent resonant

vibrations of the fibers and surrounding soft tissues [22, 35]. MMG has been used to describe

motor control strategies, force generation capabilities, to provide biofeedback, and control access

technologies. In some studies looking at optimizing muscle activity, participants showed more

ideal responses when presented with MMG than with EMG biofeedback [25, 133]. Alternatively,

muscle activity can be measured and MMG onset and offset can be used to control a musical

instrument [134, 135]. These studies support the development of MMG into interactive tools that

can be implemented into rehabilitation.

In the literature, we find several groups using mutual entrainment and interactive RAS for gait

therapy [17, 136]. Mutual entrainment occurs when a synchronized gait pattern is formed between

the user and the therapist, in this case, the “WalkMate” robot. Originally developed by Miyake

[136], “WalkMate” is an interactive auditory-cuing robot that measures and alters heel strike

auditory patterns to improve acute fractal dynamics of pathological gait. In their system, rhythmic

Page 73: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

59

sounds corresponding to the timing of footsteps are exchanged between the user and the Walk

Mate robot, showing that both rhythms adapt mutually after to each other and a stable

synchronization is automatically generated. The hierarchical control of this nonlinear model is

divided into two modules that link sensory input and motor output. The first module generates a

walking rhythm based on an artificial rhythm generator and the user’s rhythm, and then defines

step timing. This rhythmic cue embeds the auditory information into a predictable rhythm, which

allows for anticipatory movement preparation and execution. The second module adjusts the

timing difference between the sensory input and the motor output in order to converge the two

rhythms towards a target phase. This type of interactive RAS has shown short-term lasting effects

on gait patterns after a six-week period of gait training [17, 136].

Despite improvements to gait, one of the limitations with these interventions is that they are

performed in a laboratory setting, which creates a need for more accessibility in outside settings.

Additionally, the recent influx of focus on wearable technologies in community based settings

highlights the need for the integration of wearable systems into rehabilitation.

In this paper, we present the design, implementation and evaluation of a novel MMG-based

biofeedback tool, GaitTool App, for mobile phone and Smartphone application. MMG temporal

features are used to distinguish between typical and atypical gait patterns and define harmonized

musical feedback. This app will allows us to provide a rehab intervention that is easily incorporated

into daily life, while studying the long term effects of this type of training program on gait patterns

and ultimately, neural reorganization.

5.3 Mobile App Design

The core GaitTool App is based on an experimental protocol for gait intervention, MMG muscle

activity detection, Arduino processing, and gait analysis algorithms. The GaitTool App

architecture is shown below in Figure 5-1.

The fundamental framework of the mobile application is based on: a. measurement of muscle

activity via MMG; b. temporal feature detection; and c. providing users with an auditory cue during

gait.

Page 74: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

60

Figure 5-1 System flow of GaitTool app showing the main components at both the user and system

levels: MMG muscle activity measurement (A), gait analysis and feature extraction (B), and

auditory biofeedback (C).

5.3.1 Design Considerations

Intuitive Auditory Biofeedback – convert real-time MMG data to an intuitive auditory output

that distinguishes between typical and atypical gait.

User-Specific Pattern Detection – MMG features are selected based on user’s gait abilities.

Muscle Function Monitoring – tracking MMG to monitor short- and long-term changes in

muscle activity patterns and gait dynamics.

Page 75: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

61

5.3.2 MMG Muscle Activity

MMG is acquired at a sampling frequency of 1 kHz with two tri-axial accelerometers (ADXL337,

Analog Devices Inc, Norwood, MA) positioned over the tibialis anterior (TA) and lateral

gastrocnemius (LG) muscle bellies, bilaterally (Figure 5-2). Accelerometers were taped directly

onto the skin, and wired to an Arduino UNO MCU board (ATmega328-based, Arduino) with

ribbon cables.

Figure 5-2 MMG sensors taped directly onto the muscle bellies of the tibialis anterior (A) and

lateral gastrocnemius (B). In this initial prototype, the user is able to carry the Arduinos in his

pockets during gait.

To extract MMG, the z-component of the accelerometer data is filtered between 5 to 50 Hz (2nd

order Butterworth) and squared. The mean of 150 squared samples is then low-pass filtered at 5

Hz (3rd order Butterworth).

Page 76: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

62

Figure 5-3 Example of filtered MMG of TA (blue) and LG (orange) showing aligned peaks (left)

that create a harmony and misaligned peaks (right) that do not create a harmony.

5.3.3 Arduino Processing

This tool processes accelerometer data in a two-stage process. All Arduino code is written in the

C programming language. First, for every millisecond, accelerometer data is read into a 5-wide

circular buffer (b0). The buffer is passed through a filter (previously described in section ii) whose

value is stored in another 5-wide circular buffer (b1). The square of the newest value in b1 is added

to the running sum of the current 15-wide window (newMS).

Secondly, for every 15 milliseconds, the value of newMS is stored in the next index of a 10-wide

buffer of old MS values (b2), and the newMS is added to the first value of a 4-wide mean squared

buffer (b3). Once all 10 positions of b2 have been filled, the first value of b3 is set to be the mean

of the 150 collected samples by dividing by 150. This initialization period allows the system to

stabilize and ensure a consistent relation with the sensors.

After initialization, the next value in b3 is set to be the mean of the 9 last 15-sample blocks of data

from b2 plus the latest newMS value. The oldest value in b2 is then replaced with newMS. Then,

b3 is passed through a 3rd order Butterworth low-pass filter (5 Hz cutoff) and then the filtered data

Page 77: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

63

is stored in a new 4-wide buffer (b4). A two-way Bluetooth connection between the mobile app

and the BLE Mini devices is established using RedBearLab’s Bluetooth library (Red Bear

Company Limited, 2015). Bluetooth is handled by a GATT service that will run on the mobile

device, and its receiver will listen for new data from either leg. Pre-processed MMG data is

transmitted to the app every 15 ms for further analysis.

5.3.4 Gait Analysis and Feature Extraction

This tool provides rhythmic entrainment based on the temporal patterns of MMG muscle activity.

The app was developed in Android Studio (Creative Commons Attribution 2.5), using the Java

programming language.

The custom code begins by analyzing melody contours of TA and LG MMG, and defining

temporal differences between muscle peaks. Positive sounds are played when muscle peaks align,

and negative sounds are played when muscle peaks do not align (Figure 5-3). Due to the pre-

processing done by the Arduino, peaks are smooth and easily detectable on the mobile device.

Alignment is defined as when the peaks from each muscle occur within 50 ms of each other. If the

peaks are more than this but less than 250 ms apart, this is defined as misalignment. If the time

difference is any more than 250 ms, the peaks are not considered to be related, i.e., part of a

different gait cycle.

Currently, this tool compares users’ MMG against typical patterns seen in the youth and adult

MMG during self-paced gait [55]. Based on previous clinical and therapy observations, children

and adults with neurological movement disorders have variable gait patterns, which are difficult

to classify across participants [118, 137]. As such, user gait needs to be pre-analyzed and gait

features need to be pre-defined by a therapist or clinician prior to at-home use. MMG gait data will

be stored on the tablet for further long-term analysis on the effectiveness of this tool for improving

gait patterns.

5.3.5 Auditory Biofeedback

A custom drum kit was developed in PureData (Pd, puredata.info, hosted by IEM) to provide a

realistic musical cuing system. The Pd for Android library allows the app to communicate with the

Pd software. Four kits and four sounds were developed based on various musical styles and

Page 78: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

64

instruments: rock, funk, samba, 808, trombone, quacks, etc. These libraries can be personalized

for user preference.

Upon app startup, sound libraries to be played as auditory biofeedback will be selected by the user

(Figure 5-4). Auditory biofeedback will be presented to the user as a melody made up of a fixed

RAS based on the step sounds and an interactive RAS based on the drum kit layers.

Figure 5-4 Example of filtered MMG of TA (blue) and LG (orange) showing aligned peaks (left)

that create a harmony and misaligned peaks (right) that do not create a harmony.

5.3.6 Use Case

The app is designed to be intuitive and user-friendly. On startup, the user selects the desired drum

kit and sound library to move with. A switch may be selected to run the app in “Therapy Mode”,

where a therapist can control values such as thresholds to define gait health, or the RAS tempo.

Page 79: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

65

During this screen, the user will be shown the two devices that the app connects to. Once all

selections and connections are made, the user can press the “Go” button to begin.

The main window of the app will have a “Start” button for the user to select once ready to walk.

This tells both Arduinos to begin data processing through a Bluetooth transmission. Instantly, one

graph for each leg will become populated with real-time processed data from each Arduino. When

aligned and misaligned peaks occur, the user will hear each respective sound. The drum machine

will begin once the initialization process is complete, and will continue indefinitely.

5.4 Mobile App Implementation

This tool was developed based on a combination of rhythmic and mutual entrainment, thus similar

to previous interactive RAS tools [17, 136], this tool establishes mutual entrainment between the

user and the app and provides a modulated auditory cue back to the user.

First, entrainment is established by playing a fixed tempo that is predetermined by the user’s

walking abilities and therapy goals. To initiate the app, the user must walk consistently to establish

a baseline, and stride rate is calculated based on collected MMG peaks. A simple auditory cue is

presented at the pre-set tempo once 10 steps are timed to be within 2 standard deviations of each

other. Studies show that if the tempo provided by RAS is too different from spontaneous walking

tempo, the RAS presented will conflict with natural rhythm and have a negative impact on gait

dynamics [17].

Second, interactive RAS feedback is provided using temporal features selected from live MMG

measurements. In this tool, user gait patterns will be compared against typical gait patterns and

determined to be typical or atypical [55]. If gait patterns align, users will be presented with a

musical cue that sounds pleasant, whereas if patterns do not align, an unpleasant sound will be

presented. The tool will also track how many typical steps are taken, and if a given threshold is

reached, a layer of music from a programmed drum kit will be added to the auditory feedback.

Positive patterns will increase layers of the drum kit as the user takes successive steps, whereas

unsuccessful steps will need to be tallied up to remove drum kit layers. Studies have shown the

complexity of movement and locomotion [14, 16], thus we propose a dual system of fixed RAS

Page 80: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

66

will stimulate pattern synchronization and then MMG-based biofeedback for stimulating temporal

cues of movements to improve gait dynamics and neural reorganization in the long-term.

In terms of therapy protocol, users will perform 30 minutes of GaitTool App-assisted walking

daily for 8 weeks. In a recent study looking at motor recovery during early gait rehabilitation in

neurological disorders, motor deficits were seen to improve in as early as four weeks with daily

gait training [138]. This at-home tool allows us to increase length of gait training thereby

enhancing neural reorganization and lasting motor improvements [139]. Additionally, this study

will allow us to gain knowledge on optimizing training parameters for gait therapy.

5.5 App Evaluation

In this section, we discuss initial results from the evaluation of the GaitTool App prototype. After

device validation, and therapy validation, the GaitTool App will be ready for the market.

We will implement the application for Android operating system and tested it on a tablet (Samsung

Note 6 with Android 7.0 Nougat). Validation and prototype testing will occur in two stages. First,

laboratory testing will be done to evaluate the App for use with the clinical population. We are

currently recruiting participants for this study. Self-paced gait will be recorded offline and MMG

patterns will be analyzed for participants with diplegic cerebral palsy. Preliminary online testing

with feedback will then be performed in the laboratory to validate gait therapy performance.

Secondly, we will conduct a longitudinal study measuring the effects of an 8-week gait

intervention with the GaitTool App. MMG gait patterns and App use will be tracked to explore

the optimization of training parameters and in compliance of use of this tool.

5.6 Discussion and Conclusions

To our knowledge, this paper presents the first mobile application that plays auditory feedback

based on MMG muscle activity. Wearable MMG allow us to measure, analyze, and present real-

time auditory feedback in a way that is not only affordable and convenient, but also enjoyable.

Page 81: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

67

This tool also allows us to track gait dynamics longitudinally, which is important for tracking

therapy outcomes and neural reorganization.

5.7 Acknowledgments

Thanks to Ka Lun Tam and Pierre Duez for their help with hardware and software design and

implementation, and the members of the PRISM lab for their ongoing support. Special thanks to

Dr. Thaut for his expertise in movement sonification and assistance in identifying suitable auditory

feedback.

Page 82: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

68

Chapter 6

Conclusions

6.1 Summary of Contributions

Major contributions of the thesis are as follows:

1. Demonstrated the intermodal agreement between accelerometer-based MMG and surface

EMG muscle activity during pediatric gait using a PSO algorithm. Specifically, using a

combination of amplitude threshold ( ), moving window size ( ), and minimum percent

of EMG-MMG activity overlap (𝛿) between modalities, we discovered temporal alignment

(balanced accuracy in excess of 75%) of electrical and mechanical signals at the tibialis

anterior, vastus lateralis, and biceps femoris, and temporal misalignment (~50% balanced

accuracy) at the lateral gastrocnemius. These findings suggest that the relationship between

electrical and mechanical muscle activities can be more complicated in dynamic quasi-

periodic motor tasks than in simple, isolated contractions.

2. Demonstrated that spatiotemporal patterns of MMG during the gait cycle differ from those

of EMG. In contrast to isometric and isokinetic contractions, during a complex dynamic

motor task, the displacement of the muscle fascicles and the corresponding MMG signal

do not necessarily follow EMG activity. The timing and power distribution differences

between these modalities may in part be related to muscle fascicle length changes that are

unique to muscle motion during gait.

3. Demonstrated the mechanical synergies associated with lower limb MMG activity during

walking and running gait. Understanding mechanical synergies during motor tasks is useful

in detecting pathology, providing feedback about movement patterns, and in controlling

assistive technologies and robots.

4. Developed an Android-based application that presents auditory feedback based on the

alignment of MMG activity between the tibialis anterior and lateral gastrocnemius. This

application records and analyzes MMG in real-time, sets a rhythmic stimulus based on the

Page 83: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

69

user’s cadence, and controls a drum kit based on typical and atypical MMG patterns. This

is the first app to combine fixed rhythmic stimulation with sonification to gait therapy

targeted for at-home use.

6.2 Future Work

To further develop the use of MMG in dynamic motor tasks, such as gait, the following may be of

interest to future studies.

The proposed MMG segmentation method (Chapter 3) provides a framework for comparing

differing modalities based on pre-defined criteria, which can be tailored to specific target

populations or movement types. Future studies might focus on exploring MMG features in client

populations, as well as exploring different features in dynamic MMG, such as frequency features,

which can be helpful in detecting different contraction types [59]. In particular, wavelet-based

MMG analysis [59] for automatic detection of concentric and eccentric contractions during gait

merits further exploration.

The segmentation method also requires a small subset of training data, which may change for

different populations or movement types. Thus, training of the algorithm should be targeted to

each movement type. Additionally, future work should focus on creating a generalized detection

algorithm that is muscle-specific instead of participant-specific. This algorithm can be

implemented as a bimodal EMG-MMG tool for: the detection of voluntary muscle activity amid

motion artifact; the identification of pathologies in movement patterns; and, the discrimination

between various types of movements, such as walking and cycling. Literature suggests that such

multi-modal (hybrid EMG-MMG) systems can enhance activity detection beyond that achievable

with a single modality (EMG or MMG) [52, 72, 132].

The reported synergy analysis can be extended to combine EMG and MMG as complementary

modalities in a neuromechanical model since together, these signals provide an understanding of

the neural input and mechanical output of muscle contraction, coordination, and movement. A

neuromechanical model might inform the measurement of rehabilitation outcomes, and provide a

low-cost method for tracking movement, and sports performance. Future work should also

Page 84: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

70

investigate mechanical synergies in populations with gait disturbances, and extend synergy

analyses to other movements, such as reaching tasks.

In terms of the GaitTool App, future work should focus on implementing wireless MMG sensors,

testing the app with the target population, and eventually expand clinical testing to include other

neuromuscular populations. The app can be applied to different movement tasks, such as jumping,

or upper limb movements, such as reaching. Additionally, specific training programs can be built-

in to include therapy goals and therapist notes. From a therapist or clinician perspective, tools can

be added to monitor muscle activity, through mechanical synergies, for example, and track therapy

progression and neuroplasticity.

6.3 Publications

6.3.1 Journal Articles

Plewa, Katherine, Ali Samadani, and Tom Chau. "Comparing electro-and mechano-myographic

muscle activation patterns in self-paced pediatric gait." Journal of Electromyography and

Kinesiology 36 (2017): 73-80.

Plewa, Katherine, Silvia Orlandi, Ali Samadani, and Tom Chau. "A Novel Approach to

Automatically Identify Coincident Activity Between EMG and MMG Signals." Journal of

Electromyography and Kinesiology Submitted Nov 2017.

Plewa, Katherine, Silvia Orlandi, Kei Masani and Tom Chau. "Muscle Synergy Patterns of

Mechanical Activity during the Gait Cycle using MMG and EMG." Frontiers in Human

Neuroscience Submitted February 2018.

6.3.2 Conference Presentations

K. Plewa, M. Silverman, S. Orlandi, T. Chau, M. Thaut (2017). Designing a Wearable MMG-

based Mobile App for Gait Rehab. IEEE Life Sciences Conference: Sydney, Australia. December

2017.

K. Plewa, O. Paserin, T. Chau (2015). Feature Extraction in Accelerometer-Based

Page 85: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

71

Mechanomyography During Pediatric Gait. IEEE Engineering in Medicine and Biology

Conference: Milan, Italy. August 2015.

Plewa, K. T. Chau (2015). Dynamic Noise Reduction and Feature Extraction in Accelerometer-

based Mechanomyography during Pediatric Gait. IUPESM World Congress on Medical Physics

and Biomedical Engineering; Toronto, ON. June 7-12, 2015.

Page 86: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

72

References

[1] J. Perry and J. R. Davids, "Gait analysis: normal and pathological function," Journal of

Pediatric Orthopaedics, vol. 12, p. 815, 1992.

[2] K. J. Kokotilo, J. J. Eng, and A. Curt, "Reorganization and preservation of motor control

of the brain in spinal cord injury: a systematic review," Journal of neurotrauma, vol. 26,

pp. 2113-2126, 2009.

[3] A. Shumway-Cook and M. H. Woollacott, Motor control: translating research into

clinical practice: Lippincott Williams & Wilkins, 2007.

[4] O. Raineteau and M. E. Schwab, "Plasticity of motor systems after incomplete spinal

cord injury," Nature Reviews Neuroscience, vol. 2, pp. 263-273, 2001.

[5] J. Hausdorff, C. Peng, Z. Ladin, J. Y. Wei, and A. L. Goldberger, "Is walking a random

walk? Evidence for long-range correlations in stride interval of human gait," Journal of

Applied Physiology, vol. 78, pp. 349-358, 1995.

[6] N. Scafetta, D. Marchi, and B. J. West, "Understanding the complexity of human gait

dynamics," Chaos: An Interdisciplinary Journal of Nonlinear Science, vol. 19, pp.

026108-026108-10, 2009.

[7] J. Hausdorff, J. Schaafsma, Y. Balash, A. Bartels, T. Gurevich, and N. Giladi, "Impaired

regulation of stride variability in Parkinson's disease subjects with freezing of gait,"

Experimental Brain Research, vol. 149, pp. 187-194, 2003.

[8] R. A. Miller, M. H. Thaut, G. C. McIntosh, and R. R. Rice, "Components of EMG

symmetry and variability in parkinsonian and healthy elderly gait,"

Electroencephalography and Clinical Neurophysiology/Electromyography and Motor

Control, vol. 101, pp. 1-7, 1996.

[9] E. Melis, R. Torres-Moreno, H. Barbeau, and E. Lemaire, "Analysis of assisted-gait

characteristics in persons with incomplete spinal cord injury," Spinal Cord, vol. 37, pp.

430-439, 1999.

[10] E. Dursun, N. Dursun, and D. Alican, "Effects of biofeedback treatment on gait in

children with cerebral palsy," Disability & Rehabilitation, vol. 26, pp. 116-120, 2004.

[11] O. M. Giggins, U. Persson, and B. Caulfield, "Biofeedback in rehabilitation," Journal of

neuroengineering and rehabilitation, vol. 10, p. 60, 2013.

Page 87: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

73

[12] S. E. Morris, "Music and Hemi-Sync in the treatment of children with developmental

disabilities," Open Ear, vol. 2, pp. 14-17, 1996.

[13] S. Bahrami, M. A. Thomas, M. Bahrami, and A. Naghizadeh, "Neurologic Music

Therapy to Facilitate Recovery from Complications of Neurologic Diseases," Journal of

Neurology and Neuroscience, vol. 8, 2017.

[14] M. H. Thaut, "Entrainment and the motor system," Music Therapy Perspectives, vol. 31,

pp. 31-34, 2013.

[15] E. E. Kwak, "Effect of rhythmic auditory stimulation on gait performance in children

with spastic cerebral palsy," Journal of music therapy, vol. 44, pp. 198-216, 2007.

[16] M. Thaut, G. McIntosh, R. Rice, R. Miller, J. Rathbun, and J. Brault, "Rhythmic auditory

stimulation in gait training for Parkinson's disease patients," Movement disorders, vol. 11,

pp. 193-200, 1996.

[17] M. J. Hove, K. Suzuki, H. Uchitomi, S. Orimo, and Y. Miyake, "Interactive rhythmic

auditory stimulation reinstates natural 1/f timing in gait of Parkinson's patients," PLoS

one, vol. 7, p. e32600, 2012.

[18] C. P. Hurt, R. R. Rice, G. C. McIntosh, and M. H. Thaut, "Rhythmic auditory stimulation

in gait training for patients with traumatic brain injury," Journal of Music Therapy, vol.

35, pp. 228-241, 1998.

[19] E. Waters, E. Davis, G. M. Ronen, P. Rosenbaum, M. Livingston, and S. Saigal, "Quality

of life instruments for children and adolescents with neurodisabilities: how to choose the

appropriate instrument," Developmental medicine & child neurology, vol. 51, pp. 660-

669, 2009.

[20] T. W. Beck, T. J. Housh, G. O. Johnson, J. P. Weir, J. T. Cramer, J. W. Coburn, et al.,

"Comparison of Fourier and wavelet transform procedures for examining the

mechanomyographic and electromyographic frequency domain responses during

fatiguing isokinetic muscle actions of the biceps brachii," Journal of Electromyography

and Kinesiology, vol. 15, pp. 190-199, 2005.

[21] C. Orizio, M. Gobbo, B. Diemont, F. Esposito, and A. Veicsteinas, "The surface

mechanomyogram as a tool to describe the influence of fatigue on biceps brachii motor

unit activation strategy. Historical basis and novel evidence," European journal of

applied physiology, vol. 90, pp. 326-336, 2003.

Page 88: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

74

[22] C. Orizio, B. Diemont, F. Esposito, E. Alfonsi, G. Parrinello, A. Moglia, et al., "Surface

mechanomyogram reflects the changes in the mechanical properties of muscle at fatigue,"

European journal of applied physiology and occupational physiology, vol. 80, pp. 276-

284, 1999.

[23] C. Orizio, D. Liberati, C. Locatelli, D. De Grandis, and A. Veicsteinas, "Surface

mechanomyogram reflects muscle fibres twitches summation," Journal of Biomechanics,

vol. 29, pp. 475-481, 1996.

[24] C. Orizio, "Muscle sound: bases for the introduction of a mechanomyographic signal in

muscle studies," Critical reviews in biomedical engineering, vol. 21, pp. 201-243, 1992.

[25] P. Madeleine, P. Vedsted, A. K. Blangsted, G. Sjogaard, and K. Sogaard, "Effects of

electromyographic and mechanomyographic biofeedback on upper trapezius muscle

activity during standardized computer work," Ergonomics, vol. 49, pp. 921-933, 2006.

[26] K. Søgaard, A. Blangsted, L. Jørgensen, P. Madeleine, and G. Sjøgaard, "Evidence of

long term muscle fatigue following prolonged intermittent contractions based on

mechano-and electromyograms," Journal of Electromyography and Kinesiology, vol. 13,

pp. 441-450, 2003.

[27] R. McLaren, F. Joseph, C. Baguley, and D. Taylor, "A review of e-textiles in

neurological rehabilitation: How close are we?," Journal of NeuroEngineering and

Rehabilitation, vol. 13, p. 1, 2016.

[28] A. Posatskiy and T. Chau, "Design and evaluation of a novel microphone-based

mechanomyography sensor with cylindrical and conical acoustic chambers," Medical

engineering & physics, vol. 34, pp. 1184-1190, 2012.

[29] A. O. Posatskiy and T. Chau, "The effects of motion artifact on mechanomyography: A

comparative study of microphones and accelerometers," Journal of Electromyography

and Kinesiology, vol. 22, pp. 320-324, 2012.

[30] T. W. Beck, T. J. Housh, J. T. Cramer, J. P. Weir, G. O. Johnson, J. W. Coburn, et al.,

"Mechanomyographic amplitude and frequency responses during dynamic muscle

actions: a comprehensive review," Biomed Eng Online, vol. 4, p. 67, 2005.

[31] J. L. Emken and D. J. Reinkensmeyer, "Robot-enhanced motor learning: accelerating

internal model formation during locomotion by transient dynamic amplification," Neural

Systems and Rehabilitation Engineering, IEEE Transactions on, vol. 13, pp. 33-39, 2005.

Page 89: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

75

[32] E. Krueger, E. Scheeren, G. Nogueira-Neto, V. Button, and P. Nohama, "Preliminary

evaluation of mechanomyographic signal of rectus femoris muscle between spinal cord

injured and healthy subjects," 2011.

[33] E. Krueger-Beck, E. M. Scheeren, G. N. Nogueira-Neto, V. L. S. Button, and P. Nohama,

"Mechanomyographic response during FES in healthy and paraplegic subjects," in

Engineering in Medicine and Biology Society (EMBC), 2010 Annual International

Conference of the IEEE, 2010, pp. 626-629.

[34] T. Uchiyama and Y. Miyazaki, "Independent Component Analysis of Mechanomyogram

Detected with an Acceleration Sensor in Motion," in World Congress on Medical Physics

and Biomedical Engineering May 26-31, 2012, Beijing, China, 2013, pp. 461-464.

[35] M. A. Islam, K. Sundaraj, R. B. Ahmad, and N. U. Ahamed, "Mechanomyogram for

Muscle function assessment: A review," PloS one, vol. 8, p. e58902, 2013.

[36] M. O. Ibitoye, N. A. Hamzaid, J. M. Zuniga, and A. K. Abdul Wahab,

"Mechanomyography and muscle function assessment: A review of current state and

prospects," Clinical Biomechanics, 2014.

[37] L. Qi, "Use of wavelet analysis techniques with surface EMG and MMG to characterise

motor unit recruitment patterns of shoulder muscles during wheelchair propulsion and

voluntary contraction tasks," UCL (University College London), 2011.

[38] P. Madeleine and L. Arendt-Nielsen, "Experimental muscle pain increases

mechanomyographic signal activity during sub-maximal isometric contractions," Journal

of electromyography and kinesiology, vol. 15, pp. 27-36, 2005.

[39] N. Alves and T. Chau, "Automatic detection of muscle activity from mechanomyogram

signals: a comparison of amplitude and wavelet-based methods," Physiological

measurement, vol. 31, p. 461, 2010.

[40] N. Alves and T. Chau, "Research The design and testing of a novel mechanomyogram-

driven switch controlled by small eyebrow movements," 2010.

[41] E. Krueger, E. M. Scheeren, G. N. Nogueira-Neto, V. L. d. S. N. Button, and P. Nohama,

"Advances and perspectives of mechanomyography," Revista Brasileira de Engenharia

Biomédica, vol. 30, pp. 384-401, 2014.

[42] M. S. Hossain, K. Sundaraj, L. C. Kiang, and Z. Said, "Mechanomyography based

muscle movement analysis: a brief," MoHE 2014, 2014.

Page 90: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

76

[43] E. Cè, S. Rampichini, and F. Esposito, "Novel insights into skeletal muscle function by

mechanomyography: from the laboratory to the field," Sport Sciences for Health, vol. 11,

pp. 1-28, 2015.

[44] K. Momen, S. Krishnan, and T. Chau, "Real-time classification of forearm

electromyographic signals corresponding to user-selected intentional movements for

multifunction prosthesis control," IEEE Transactions on Neural Systems and

Rehabilitation Engineering, vol. 15, pp. 535-542, 2007.

[45] M. Karg, A.-A. Samadani, R. Gorbet, K. Kühnlenz, J. Hoey, and D. Kulić, "Body

movements for affective expression: A survey of automatic recognition and generation,"

IEEE Transactions on Affective Computing, vol. 4, pp. 341-359, 2013.

[46] E. Mitchell, A. Ahmadi, N. E. O'Connor, C. Richter, E. Farrell, J. Kavanagh, et al.,

"Automatically detecting asymmetric running using time and frequency domain

features," in 2015 IEEE 12th International Conference on Wearable and Implantable

Body Sensor Networks (BSN), 2015, pp. 1-6.

[47] Y. Nolan, "The mechanomyogram as a channel of communication and control for the

disabled," in Engineering in Medicine and Biology Society, 2004. IEMBS'04. 26th Annual

International Conference of the IEEE, 2004, pp. 4928-4931.

[48] H. Yoshimi, K. Sasaguri, K. Tamaki, and S. Sato, "Identification of the occurrence and

pattern of masseter muscle activities during sleep using EMG and accelerometer

systems," Head & face medicine, vol. 5, p. 7, 2009.

[49] N. Alves and T. Chau, "Uncovering patterns of forearm muscle activity using multi-

channel mechanomyography," Journal of Electromyography and Kinesiology, vol. 20,

pp. 777-786, 2010.

[50] D. T. Barry, K. E. Gordon, and G. G. Hinton, "Acoustic and surface EMG diagnosis of

pediatric muscle disease," Muscle & nerve, vol. 13, pp. 286-290, 1990.

[51] M. Reaz, M. Hussain, and F. Mohd-Yasin, "Techniques of EMG signal analysis:

detection, processing, classification and applications," Biological procedures online, vol.

8, pp. 11-35, 2006.

[52] X. Zhang, X. Chen, W.-h. Wang, J.-h. Yang, V. Lantz, and K.-q. Wang, "Hand gesture

recognition and virtual game control based on 3D accelerometer and EMG sensors," in

Proceedings of the 14th international conference on Intelligent user interfaces, 2009, pp.

401-406.

Page 91: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

77

[53] H. Begovic, G.-Q. Zhou, T. Li, Y. Wang, and Y.-P. Zheng, "Detection of the

electromechanical delay and its components during voluntary isometric contraction of the

quadriceps femoris muscle," Frontiers in physiology, vol. 5, 2014.

[54] P. Prociow, A. Wolczowski, T. G. Amaral, O. P. Dias, and J. Filipe, "Identification of

Hand Movements based on MMG and EMG Signals," in BIOSIGNALS (2), 2008, pp.

534-539.

[55] K. Plewa, A. Samadani, and T. Chau, "Comparing electro-and mechano-myographic

muscle activation patterns in self-paced pediatric gait," Journal of Electromyography and

Kinesiology, vol. 36, pp. 73-80, 2017.

[56] L. Vaisman, J. Zariffa, and M. R. Popovic, "Application of singular spectrum-based

change-point analysis to EMG-onset detection," Journal of Electromyography and

Kinesiology, vol. 20, pp. 750-760, 2010.

[57] J. F.-S. Lin, A.-A. Samadani, and D. Kulić, "Segmentation by Data Point Classification

Applied to Forearm Surface EMG," in Smart City 360°, 2016, pp. 153-165.

[58] D. Bratton and J. Kennedy, "Defining a standard for particle swarm optimization," in

Swarm Intelligence Symposium, 2007. SIS 2007. IEEE, 2007, pp. 120-127.

[59] T. W. Beck, V. von Tscharner, T. J. Housh, J. T. Cramer, J. P. Weir, M. H. Malek, et al.,

"Time/frequency events of surface mechanomyographic signals resolved by nonlinearly

scaled wavelets," Biomedical Signal Processing and Control, vol. 3, pp. 255-266, 2008.

[60] A. Subasi, "Classification of EMG signals using PSO optimized SVM for diagnosis of

neuromuscular disorders," Computers in biology and medicine, vol. 43, pp. 576-586,

2013.

[61] B. Whittington, A. Silder, B. Heiderscheit, and D. G. Thelen, "The contribution of

passive-elastic mechanisms to lower extremity joint kinetics during human walking,"

Gait & posture, vol. 27, pp. 628-634, 2008.

[62] E. Clancy, E. Morin, and R. Merletti, "Sampling, noise-reduction and amplitude

estimation issues in surface electromyography," Journal of electromyography and

kinesiology: official journal of the International Society of Electrophysiological

Kinesiology, vol. 12, p. 1, 2002.

[63] P. Madeleine, P. Bajaj, K. Søgaard, and L. Arendt-Nielsen, "Mechanomyography and

electromyography force relationships during concentric, isometric and eccentric

contractions," Journal of electromyography and kinesiology, vol. 11, pp. 113-121, 2001.

Page 92: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

78

[64] J. M. Hidler and A. E. Wall, "Alterations in muscle activation patterns during robotic-

assisted walking," Clinical Biomechanics, vol. 20, pp. 184-193, 2005.

[65] N. Lythgo, C. Wilson, and M. Galea, "Basic gait and symmetry measures for primary

school-aged children and young adults whilst walking barefoot and with shoes," Gait &

posture, vol. 30, pp. 502-506, 2009.

[66] V. Von Tscharner and B. Goepfert, "Estimation of the interplay between groups of fast

and slow muscle fibers of the tibialis anterior and gastrocnemius muscle while running,"

Journal of Electromyography and Kinesiology, vol. 16, pp. 188-197, 2006.

[67] A. Den Otter, A. Geurts, T. Mulder, and J. Duysens, "Abnormalities in the temporal

patterning of lower extremity muscle activity in hemiparetic gait," Gait & posture, vol.

25, pp. 342-352, 2007.

[68] G. S. Chleboun, A. B. Busic, K. K. Graham, and H. A. Stuckey, "Fascicle length change

of the human tibialis anterior and vastus lateralis during walking," journal of orthopaedic

& sports physical therapy, vol. 37, pp. 372-379, 2007.

[69] S. M. Marek, J. T. Cramer, A. L. Fincher, L. L. Massey, S. M. Dangelmaier, S.

Purkayastha, et al., "Acute effects of static and proprioceptive neuromuscular facilitation

stretching on muscle strength and power output," Journal of Athletic Training, vol. 40, p.

94, 2005.

[70] A. P. Harrison, B. Danneskiold‐ Samsøe, and E. M. Bartels, "Portable acoustic

myography–a realistic noninvasive method for assessment of muscle activity and

coordination in human subjects in most home and sports settings," Physiological reports,

vol. 1, p. e00029, 2013.

[71] L. Qi, J. M. Wakeling, A. Green, K. Lambrecht, and M. Ferguson-Pell, "Spectral

properties of electromyographic and mechanomyographic signals during isometric ramp

and step contractions in biceps brachii," Journal of Electromyography and Kinesiology,

vol. 21, pp. 128-135, 2011.

[72] J.-Y. Guo, Y.-P. Zheng, H.-B. Xie, and X. Chen, "Continuous monitoring of

electromyography (EMG), mechanomyography (MMG), sonomyography (SMG) and

torque output during ramp and step isometric contractions," Medical engineering &

physics, vol. 32, pp. 1032-1042, 2010.

[73] E. Azizi, E. L. Brainerd, and T. J. Roberts, "Variable gearing in pennate muscles,"

Proceedings of the National Academy of Sciences, vol. 105, pp. 1745-1750, 2008.

Page 93: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

79

[74] K. Kubo, H. Kanehisa, K. Azuma, M. Ishizu, S.-Y. Kuno, M. Okada, et al., "Muscle

architectural characteristics in women aged 20-79 years," Medicine and science in sports

and exercise, vol. 35, pp. 39-44, 2003.

[75] M. Ishikawa, P. V. Komi, M. J. Grey, V. Lepola, and G.-P. Bruggemann, "Muscle-tendon

interaction and elastic energy usage in human walking," Journal of applied physiology,

vol. 99, pp. 603-608, 2005.

[76] M. Ishikawa, J. Pakaslahti, and P. Komi, "Medial gastrocnemius muscle behavior during

human running and walking," Gait & posture, vol. 25, pp. 380-384, 2007.

[77] T. Fukunaga, Y. Kawakami, K. Kubo, and H. Kanehisa, "Muscle and tendon interaction

during human movements," Exercise and sport sciences reviews, vol. 30, pp. 106-110,

2002.

[78] T. Fukunaga, K. Kubo, Y. Kawakami, S. Fukashiro, H. Kanehisa, and C. N. Maganaris,

"In vivo behaviour of human muscle tendon during walking," Proceedings of the Royal

Society of London B: Biological Sciences, vol. 268, pp. 229-233, 2001.

[79] G. Lichtwark, K. Bougoulias, and A. Wilson, "Muscle fascicle and series elastic element

length changes along the length of the human gastrocnemius during walking and

running," Journal of biomechanics, vol. 40, pp. 157-164, 2007.

[80] M. E. Héroux, C. J. Dakin, B. L. Luu, J. T. Inglis, and J.-S. Blouin, "Absence of lateral

gastrocnemius activity and differential motor unit behavior in soleus and medial

gastrocnemius during standing balance," Journal of applied physiology, vol. 116, pp.

140-148, 2014.

[81] D. Steins, H. Dawes, P. Esser, and J. Collett, "Wearable accelerometry-based technology

capable of assessing functional activities in neurological populations in community

settings: a systematic review," Journal of neuroengineering and rehabilitation, vol. 11, p.

1, 2014.

[82] O. Mazumder and A. S. Kundu, "EMG Based Multichannel Human Computer Interface

for Rehabilitation Training," in 8th National Conference on Medical Informatics, 2012.

[83] R. Merletti and D. Farina, Surface electromyography: physiology, engineering and

applications: John Wiley & Sons, 2016.

[84] J. T. Blackburn, D. R. Bell, M. F. Norcross, J. D. Hudson, and L. A. Engstrom,

"Comparison of hamstring neuromechanical properties between healthy males and

Page 94: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

80

females and the influence of musculotendinous stiffness," Journal of Electromyography

and Kinesiology, vol. 19, pp. e362-e369, 2009.

[85] Ş. U. Yavuz, A. Şendemir-Ürkmez, and K. S. Türker, "Effect of gender, age, fatigue and

contraction level on electromechanical delay," Clinical Neurophysiology, vol. 121, pp.

1700-1706, 2010.

[86] A. Jaskólska, W. Brzenczek, K. Kisiel-Sajewicz, A. Kawczyński, J. Marusiak, and A.

Jaskólski, "The effect of skinfold on frequency of human muscle mechanomyogram,"

Journal of Electromyography and Kinesiology, vol. 14, pp. 217-225, 2004.

[87] S. R. Perry-Rana, T. J. Housh, G. O. Johnson, A. J. Bull, and J. T. Cramer, "MMG and

EMG responses during 25 maximal, eccentric, isokinetic muscle actions," Medicine and

science in sports and exercise, vol. 35, pp. 2048-2054, 2003.

[88] S. R. Perry, T. J. Housh, J. P. Weir, G. O. Johnson, A. J. Bull, and K. T. Ebersole, "Mean

power frequency and amplitude of the mechanomyographic and electromyographic

signals during incremental cycle ergometry," Journal of Electromyography and

Kinesiology, vol. 11, pp. 299-305, 2001.

[89] Y. Yoshitake and T. Moritani, "The muscle sound properties of different muscle fiber

types during voluntary and electrically induced contractions," Journal of

Electromyography and Kinesiology, vol. 9, pp. 209-217, 1999.

[90] Y. Yoshitake, M. Shinohara, H. Ue, and T. Moritani, "Characteristics of surface

mechanomyogram are dependent on development of fusion of motor units in humans,"

Journal of Applied Physiology, vol. 93, pp. 1744-1752, 2002.

[91] N. Matsuhisa, M. Kaltenbrunner, T. Yokota, H. Jinno, K. Kuribara, T. Sekitani, et al.,

"Printable elastic conductors with a high conductivity for electronic textile applications,"

Nature communications, vol. 6, 2015.

[92] J. Marusiak, A. Jaskólska, K. Kisiel-Sajewicz, G. H. Yue, and A. Jaskólski, "EMG and

MMG activities of agonist and antagonist muscles in Parkinson’s disease patients during

absolute submaximal load holding," Journal of electromyography and kinesiology, vol.

19, pp. 903-914, 2009.

[93] J. Marusiak, A. Jaskólska, E. Jarocka, W. Najwer, K. Kisiel‐ Sajewicz, and A. Jaskólski,

"Electromyography and mechanomyography of elbow agonists and antagonists in

Parkinson disease," Muscle & nerve, vol. 40, pp. 240-248, 2009.

Page 95: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

81

[94] D. Sutherland, "The development of mature gait," Gait & Posture, vol. 6, pp. 163-170,

1997.

[95] E. Sejdić, Y. Fu, A. Pak, J. A. Fairley, and T. Chau, "The effects of rhythmic sensory

cues on the temporal dynamics of human gait," PloS one, vol. 7, p. e43104, 2012.

[96] M. Shinohara, M. Kouzaki, T. Yoshihisa, and T. Fukunaga, "Mechanomyography of the

human quadriceps muscle during incremental cycle ergometry," European journal of

applied physiology and occupational physiology, vol. 76, pp. 314-319, 1997.

[97] A. Jaskólski, R. Andrzejewska, J. Marusiak, K. Kisiel-Sajewicz, and A. Jaskólska,

"Similar response of agonist and antagonist muscles after eccentric exercise revealed by

electromyography and mechanomyography," Journal of Electromyography and

Kinesiology, vol. 17, pp. 568-577, 2007.

[98] K. Hase and R. Stein, "Turning strategies during human walking," Journal of

Neurophysiology, vol. 81, pp. 2914-2922, 1999.

[99] G. Cappellini, Y. P. Ivanenko, R. E. Poppele, and F. Lacquaniti, "Motor patterns in

human walking and running," Journal of neurophysiology, vol. 95, pp. 3426-3437, 2006.

[100] Y. P. Ivanenko, R. E. Poppele, and F. Lacquaniti, "Five basic muscle activation patterns

account for muscle activity during human locomotion," The Journal of physiology, vol.

556, pp. 267-282, 2004.

[101] K. P. Granata, D. A. Padua, and M. F. Abel, "Repeatability of surface EMG during gait in

children," Gait & posture, vol. 22, pp. 346-350, 2005.

[102] Y. A. Koryak, "Functional and clinical significance of the architecture of human skeletal

muscles," Human Physiology, vol. 34, pp. 482-492, 2008.

[103] E. Cè, S. Longo, S. Rampichini, M. Devoto, E. Limonta, M. Venturelli, et al., "Stretch-

induced changes in tension generation process and stiffness are not accompanied by

alterations in muscle architecture of the middle and distal portions of the two

gastrocnemii," Journal of Electromyography and Kinesiology, vol. 25, pp. 469-478,

2015.

[104] A. Jaskólska, P. Madeleine, A. Jaskólski, K. Kisiel-Sajewicz, and L. Arendt-Nielsen, "A

comparison between mechanomyographic condenser microphone and accelerometer

measurements during submaximal isometric, concentric and eccentric contractions,"

Journal of Electromyography and Kinesiology, vol. 17, pp. 336-347, 2007.

Page 96: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

82

[105] V. Linnamo, V. Strojnik, and P. Komi, "EMG power spectrum and features of the

superimposed M-wave during voluntary eccentric and concentric actions at different

activation levels," European journal of applied physiology, vol. 86, pp. 534-540, 2002.

[106] L. H. Ting and J. L. McKay, "Neuromechanics of muscle synergies for posture and

movement," Current opinion in neurobiology, vol. 17, pp. 622-628, 2007.

[107] E. Bizzi and V. C. Cheung, "The neural origin of muscle synergies," Frontiers in

computational neuroscience, vol. 7, p. 51, 2013.

[108] S. Hagio, M. Fukuda, and M. Kouzaki, "Identification of muscle synergies associated

with gait transition in humans," Frontiers in human neuroscience, vol. 9, p. 48, 2015.

[109] R. R. Neptune, K. Sasaki, and S. A. Kautz, "The effect of walking speed on muscle

function and mechanical energetics," Gait & posture, vol. 28, pp. 135-143, 2008.

[110] L. H. Ting, S. A. Chvatal, S. A. Safavynia, and J. Lucas McKay, "Review and

perspective: neuromechanical considerations for predicting muscle activation patterns for

movement," International journal for numerical methods in biomedical engineering, vol.

28, pp. 1003-1014, 2012.

[111] S. Aoi and T. Funato, "Neuromusculoskeletal models based on the muscle synergy

hypothesis for the investigation of adaptive motor control in locomotion via sensory-

motor coordination," Neuroscience research, vol. 104, pp. 88-95, 2016.

[112] B. Ghali, N. T. Anantha, J. Chan, and T. Chau, "Variability of grip kinetics during adult

signature writing," PLoS One, vol. 8, p. e63216, 2013.

[113] A. H. Huntley, J. L. Zettel, and L. A. Vallis, "Older adults exhibit altered motor

coordination during an upper limb object transport task requiring a lateral change in

support," Human movement science, vol. 52, pp. 133-142, 2017.

[114] E. C. Hill, T. J. Housh, C. M. Smith, K. C. Cochrane, N. D. Jenkins, R. J. Schmidt, et al.,

"The Effects of Work-to-Rest Ratios on Torque, Electromyographic, and

Mechanomyographic Responses to Fatiguing Workbouts," International Journal of

Exercise Science, vol. 10, pp. 580-591, 2017.

[115] E. Bichler and J. Celichowski, "Changes in the properties of mechanomyographic signals

and in the tension during the fatigue test of rat medial gastrocnemius muscle motor

units," Journal of Electromyography and Kinesiology, vol. 11, pp. 387-394, 2001.

Page 97: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

83

[116] A. Jaskólska, K. Kisiel, W. Brzenczek, and A. Jaskólski, "EMG and MMG of synergists

and antagonists during relaxation at three joint angles," European journal of applied

physiology, vol. 90, pp. 58-68, 2003.

[117] K. M. Steele, M. C. Tresch, and E. J. Perreault, "The number and choice of muscles

impact the results of muscle synergy analyses," Frontiers in computational neuroscience,

vol. 7, 2013.

[118] L. Tang, F. Li, S. Cao, X. Zhang, D. Wu, and X. Chen, "Muscle synergy analysis in

children with cerebral palsy," Journal of neural engineering, vol. 12, p. 046017, 2015.

[119] E. Zwaan, J. G. Becher, and J. Harlaar, "Synergy of EMG patterns in gait as an objective

measure of muscle selectivity in children with spastic cerebral palsy," Gait & posture,

vol. 35, pp. 111-115, 2012.

[120] S. Hagio and M. Kouzaki, "The flexible recruitment of muscle synergies depends on the

required force-generating capability," Journal of neurophysiology, vol. 112, pp. 316-327,

2014.

[121] D. D. Lee and H. S. Seung, "Learning the parts of objects by non-negative matrix

factorization," Nature, vol. 401, pp. 788-791, 1999.

[122] L. H. Ting and S. A. Chvatal, "Decomposing muscle activity in motor tasks," Motor

Control Theories, Experiments and Applications. Oxf. Univ. Press, New York, pp. 102v-

138, 2010.

[123] K. Nishida, S. Hagio, B. Kibushi, T. Moritani, and M. Kouzaki, "Comparison of muscle

synergies for running between different foot strike patterns," PloS one, vol. 12, p.

e0171535, 2017.

[124] A. Hof, H. Elzinga, W. Grimmius, and J. Halbertsma, "Speed dependence of averaged

EMG profiles in walking," Gait & posture, vol. 16, pp. 78-86, 2002.

[125] K. M. Steele, A. Rozumalski, and M. H. Schwartz, "Muscle synergies and complexity of

neuromuscular control during gait in cerebral palsy," Developmental Medicine & Child

Neurology, vol. 57, pp. 1176-1182, 2015.

[126] J.-R. Lacour and M. Bourdin, "Factors affecting the energy cost of level running at

submaximal speed," European journal of applied physiology, vol. 115, pp. 651-673,

2015.

[127] A. Nummela, T. Keränen, and L. Mikkelsson, "Factors related to top running speed and

economy," International journal of sports medicine, vol. 28, pp. 655-661, 2007.

Page 98: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

84

[128] G. Lichtwark and A. Wilson, "Optimal muscle fascicle length and tendon stiffness for

maximising gastrocnemius efficiency during human walking and running," Journal of

theoretical biology, vol. 252, pp. 662-673, 2008.

[129] V. C. Cheung, K. Devarajan, G. Severini, A. Turolla, and P. Bonato, "Decomposing time

series data by a non-negative matrix factorization algorithm with temporally constrained

coefficients," in Engineering in Medicine and Biology Society (EMBC), 2015 37th

Annual International Conference of the IEEE, 2015, pp. 3496-3499.

[130] Y. Ohta, N. Shima, and K. Yabe, "In vivo behaviour of human muscle architecture and

mechanomyographic response using the interpolated twitch technique," Journal of

Electromyography and Kinesiology, vol. 19, pp. e154-e161, 2009.

[131] N. Miyamoto and S. Oda, "Mechanomyographic and electromyographic responses of the

triceps surae during maximal voluntary contractions," Journal of Electromyography and

Kinesiology, vol. 13, pp. 451-459, 2003.

[132] T. M. Vieira, R. Merletti, and L. Mesin, "Automatic segmentation of surface EMG

images: Improving the estimation of neuromuscular activity," Journal of biomechanics,

vol. 43, pp. 2149-2158, 2010.

[133] T. K. Evetovich, D. S. Conley, J. B. Todd, D. C. Rogers, and T. L. Stone, "Effect of

mechanomyography as a biofeedback method to enhance muscle relaxation and

performance," The Journal of Strength & Conditioning Research, vol. 21, pp. 96-99,

2007.

[134] M. Donnarumma, B. Caramiaux, and A. Tanaka, "Muscular Interactions Combining

EMG and MMG sensing for musical practice," 2013.

[135] M. Tanaka, T. Okuyama, and K. Saito, "Study on evaluation of muscle conditions using a

mechanomyogram sensor," in Systems, Man, and Cybernetics (SMC), 2011 IEEE

International Conference on, 2011, pp. 741-745.

[136] Y. Miyake, "Interpersonal synchronization of body motion and the Walk-Mate walking

support robot," Robotics, IEEE Transactions on, vol. 25, pp. 638-644, 2009.

[137] D. C. Johnson, D. L. Damiano, and M. F. Abel, "The evolution of gait in childhood and

adolescent cerebral palsy," Journal of Pediatric Orthopaedics, vol. 17, pp. 392-396,

1997.

[138] H. Kumru, N. Murillo, J. Benito-Penalva, J. M. Tormos, and J. Vidal, "Transcranial

direct current stimulation is not effective in the motor strength and gait recovery

Page 99: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

85

following motor incomplete spinal cord injury during Lokomat® gait training,"

Neuroscience letters, vol. 620, pp. 143-147, 2016.

[139] A. Meyer-Heim, C. Ammann-Reiffer, A. Schmartz, J. Schaefer, F. H. Sennhauser, F.

Heinen, et al., "Improvement of walking abilities after robotic-assisted locomotion

training in children with cerebral palsy," Archives of disease in childhood, 2009.

Page 100: Analysis and Interpretation of Lower Limb ... · EMG activity was >75% for all muscles sites except the gastrocnemius, where concurrent activity was only observed about 55% of the

86

*small fist pump*