cluster analysis of fmri data using dendrogram sharpening l. stanberry, r. nandy, and d. cordes...

28
Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

Upload: xiomara-satterthwaite

Post on 01-Apr-2015

222 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

Cluster Analysis of fMRI Data Using Dendrogram

Sharpening

L. Stanberry, R. Nandy, and D. Cordes

Presenter: Abdullah-Al Mahmood

Page 2: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

Outline

Problem DefinitionThe Solution

Choice of methods, parameters etc.Algorithm – Dendrogram Sharpening

Experiments and ResultsDiscussion

Page 3: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

The Task

Identify areas of activation in the brain in response to certain stimuli

Page 4: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

The Task

Identify areas of activation in the brain in response to certain stimuli

Simple case: Single StimulusPaced motor paradigm (finger

tapping)Region of Interest: Motor cortex

(Motion controlling area)

Page 5: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

The Task

Identify areas of activation in the brain in response to certain stimuli

Simple case: Single StimulusPaced motor paradigm (finger

tapping)Region of Interest: Motor cortex

(Motion controlling area)Challenges: Noise & Data Volume

Page 6: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

Basic Algorithm

Hierarchical clustering

Page 7: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

Basic Algorithm

Hierarchical clusteringFactors to consider

The (dis)similarity measureThe linkage methodThreshold for cutting tree vs. number of nodes

Page 8: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

Distance Measure

Two voxels are similar if the activation patterns are similarCorrelation coefficient of the time courses measures similarityDistance between voxels i and j

d(i, j ) = 1 – corr. coeff.(T (i ), T ( j ))

Not a metric

Page 9: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

Linkage Methods

Single – distance between closest pair of points of two clustersAverage – average distance of all pairs of points, one from each clusterComplete – largest distance between two points in two clustersSingle linkage is used in this work

Page 10: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

Single Linkage Dendrogram (SLD)

ProsCorrectly identifies structure when clusters overlapInvariant under reordering of objectsComputationally simple

Cons“Chaining effect” – highly dissimilar size of children nodes

Page 11: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

Dendrogram Example - I

Page 12: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

Dendrogram Example - II

Page 13: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

Dendrogram Sharpening

Removes chaining effect and reveals “interesting” structureDiscards some points in the process that are attached to clusters laterTwo parameters

ncore for a node/cluster (large value)

nfluff for its children (small value)

Page 14: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

Dendrogram Sharpening

The Basic AlgorithmForm a queue of nodes (initially containing root cluster only)While not empty(queue) dequeue node

If size(node) < ncore discard all points under it.Else discard child(ren) with size < nfluff and queue the remaining child(ren).

Page 15: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

Sharpening Example - I

Page 16: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

Sharpening Example - II

Page 17: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

Cluster Identification

Method of inconsistent edgesMeasure of inconsistencyThreshold = Median + 2(Upper-hinge

value – Lower-hinge value)

Upper and lower values correspond to first and third quartile values (ascending order sort for distance)

Page 18: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

Experimental Parameters

Paradigm I4 slices, each of 6464 resolution, 750 time points

Paradigm 220 slices, each of 6464 resolution, 165 time points

Activity and rest period alternated

Page 19: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

Data reduction

Discard voxels with SNR value (= mean signal intensity standard deviation) in the first decileDiscard voxels with correlation value below 0.5 (normalized series with mean 0 and std. dev. = 1) or having less than 5 significant correlations

Page 20: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

Once Sharpened Data (P – I)

Page 21: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

Twice Sharpened Data (P – I)

Page 22: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

Final classification (P – I)

Page 23: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

Map from SPM analysis

Page 24: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

A cluster from Paradigm II

Page 25: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

Numerical Comparison

Page 26: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

Discussion

Dendrogram sharpening can help in identifying clusters quite wellCan be applied to raw data as well as preprocessed dataNot tested for weak/multiple stimuliNeeds parameter tuning for sharpening algorithm

Page 27: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

Reference

L. Stanberry, R. Nandy and D. Cordes Cluster Analysis of fMRI Data Using Dendrogram Sharpening. Human Brain Mapping, 20:201-219, 2003.

N.B. All figures and tables are taken from the original work

Page 28: Cluster Analysis of fMRI Data Using Dendrogram Sharpening L. Stanberry, R. Nandy, and D. Cordes Presenter: Abdullah-Al Mahmood

Questions?