kap 8 · 4/5/2011 1 kap 8 image quality, signal, contrast and noise fys-kjm 4740 mr-teori og...
TRANSCRIPT
4/5/2011
1
Kap 8Image quality, signal, contrast and noise
FYS-KJM 4740
MR-teori og medisinsk diagnostikk
contrast SNR
speedresolution
Available MR-
parameters
contrast SNR
speedresolution
Available MR-
parameters
Main source of noise in MRI:
•Noise generated within the reciever RF electronics
•Brownian motion of electrons within the body (in
conducting tissue)
)())(exp()()( 0 tndrrtjkrMNtSv
+= ∫ω
Signal induced in received coil with N turns
(ignoring effect of sequence parameters)
n(t)=complex noise term
N=number of turns in receiver coil
NB:2
0
2
000 (t) so / ωγω ∝∝=∝−∝ BSBsignalMRM
Noise-independent signal from a voxel of volume Vh
(Macovski A. Magn Reson Med 1996)
N=No of coil terms, χ=tissue susceptibility, k=Boltzmanns const, T=temp, Tr=read-out time (time to record echo), R=coil resistance,
Signal-Noise ratio (single voxel, one measurement)
r
h
TRkT
VNSNR
/2
2
0
⋅=
χω
γχω /2
0 hsig VNM = M(r)=χB0
Does SNR scale with Bo2 ?
(probably not in reality)
Coil resistance R, complex function of Bo
SNR also function of sequence parameters and Q-factor of
coil (Q=ωL/R)
)T2*, T2,T1,,TE,S(TR,
3
0ρα⋅
⋅⋅= h
sySAV
BW
NNNQBASNR
A=constant (susceptibility, temp, object geometry, size etc)
BW=pixel bandwidth=1/Tr
NSA=number of averages
Ny=number of phase encoding steps; Ns=number of slice enc
steps (=1 for 2D)
4/5/2011
2
receiver bandwidth
measured signal
includes also noise
timedwellsamplesofnumber
samplesofnumber
echooftimesampling
samplesofnumberbandwidthreceiver
⋅==
sampling time of echo
Pixel bandwdth (receiver bandwidth):
increasing the bandwidth
sampling time of echo
receiver bandwidth sampling time
receiver bandwidth and gradient strength
frequency
frequency encoding gradient
rbw
frequency
encoding
gradient
receiver
bandwidth
faster rephasing due to gradient; i.e. echo forms faster
noise and frequency encoding
frequency/
kHz
noise
-16 16-24 24
head
0
total amount of signal of signal per voxel is unchanged,
bigger difference of resonance frequencies inside the voxel
but : total amount of noise is increased !!!
Signal vs contrast
σ
)()( BSASSNRSNRCNR BA
−=−=
σ = image noise (assumed position independent)
0 25 50 75 1000
0.05
0.1
0.15
0.2
0.25
T1=900T1=300
CNR
Flip angle (deg)
Rel
SI,
Rel C
NR
0 25 50 75 1000
0.05
0.1
0.15
0.2
0.25
T1=900T1=300
CNR
Flip angle (deg)
Rel
SI,
Rel C
NR
Signal vs contrast (T1.GRE)
4/5/2011
3
0 200 400 600 800 10000
0.5
1
1.5
T1=900
T1=300
CNR
TR (ms)
Rel
sig
nal
, re
l C
NR
0 200 400 600 800 10000
0.5
1
1.5
T1=900
T1=300
CNR
TR (ms)
Rel
sig
nal
, re
l C
NR
Signal vs contrast (SE)
)1/1ln(11
1121
21
21 TTTT
TTTR
opt⋅
−=
0 200 400 600 800 10000
0.5
1
1.5
T1=900
T1=300
CNR
TR (ms)
Rel
sig
nal
, re
l C
NR
0 200 400 600 800 10000
0.5
1
1.5
T1=900
T1=300
CNR
TR (ms)
Rel
sig
nal
, re
l C
NR
Sa Sb
σ2
Sa Sb
σ2
Practical measurement of SNR
0
5
10
15
20
25
0 5 10 15 20 25
Number of averages
Effect of NSA on SNR
Effect of BW on SNR
BW=750 Hz BW=2055 Hz
kx
ky
p1
p*1
p2
p*2
A B
Partial k-space sampling
4/5/2011
4
kx
ky
‘Half-scan’ (partial Fourier) phase partial fourier
example: half fourier
phase error due to
inhomogeneity of B0
8 further steps required
for phase correction
FOVx,y : 256 mm
matrix= 256 x (128+8)
pixel size: (1 x 1) mm2
acquisition time : ca. 50 % (256 x 256)
SNR (relative) = 1./1.41
α
TE
α
Tacq
TE
Tacq
Gx
‘Partial echo’ky
p1
kyky
Rectangular field of view (rFoV)
)T2*, T2,T1,,TE,S(TR,
3
0ρα⋅
⋅⋅=
h
sySAV
BW
NNNQBASNR
Working examples:
Ref scan: FoV 256 x 256; BW=1000 Hz/px; V=1x1x1 mm3
1. 75% rFoV
2. Half-scan (partial fourier)
3. Reduced phase sampling (reduced Ny)
4. Reduced BW =BW/2
5. Increased TR (SE sequence)
6. Increased FA (GRE sequence)
7. Partial echo
Parallel imaging
Basic concepts:
•Many small coils give better overall SNR than one large
•Phased array technology: parallel processing of signal from
multiple coils
•Requires fast data handling and high processing capacity due to
parallel processing of multiple data streams
•Use of signal sensitivity profile from each coil element to
reconstruct corrected ’undersampled’ image
4/5/2011
5
From Larkman & Nunes; Phys Med Biol (2007)
Use of multiple receive coils
From Larkman & Nunes; Phys Med Biol (2007)
Reduced K-space sampling
Sensitivity Encoding (SENSE)
•Extent of k-space (resolution ) unchanged
•Distance between adjacent k-space lines increased by factor r
•Results in signal components from r locations overlap in the
(undersampled) image
•Provided the coil sensitivty is different for each coil element,
the correct signal distribution in the whole image can be
reconstrcuted from the aliased image if the coil sensitivity
profile for each coil element is known.
The concept of parallel imaging
From Larkman & Nunes; Phys Med Biol (2007)
SENSE
S(x,y) =SI in the sub-sampled (alisased) image
C(x,y) = coil sensitivities at the location of the two aliased
pixels
ρ = signal (spin density) from the object at the two
locations (x,y) and (x, y+FoV/2)
The concept of parallel imaging
From Larkman & Nunes; Phys Med Biol (2007)
4/5/2011
6
SENSE
In matrix notation:
=>
Where ψ=receive noise matrix
From Larkman & Nunes; Phys Med Biol (2007)
SENSE does not affect spatial resolution but reduced
SNR:
r= SENSE factor
g=’g-factor’ and is a function of coil geometry, noise profile and r
r=3r=2 r=4
’g-factor’ images
From Larkman & Nunes; Phys Med Biol (2007)
SOS = sum of squares
Extraction of coil sensitivity data:
From Larkman & Nunes; Phys Med Biol (2007)
Applications of SENSE (and similar PI techniques):
•Reduced scan-time (all sequence types)
•Reduced geometric distortion and signal loss in
EPI sequences
r=1 (12 min) r=2 (6 min) r=3 (4 min)
Phase contrast angiography (PCA)
From Larkman & Nunes; Phys Med Biol (2007)
4/5/2011
7
ky
kx
0
AB
C D
EF
e1
e2
A
RF
Gx
Gy
e1
e2
B C D E F
ky
kx
0
AB
C D
EF
e1
e2
ky
kx
0
AB
C D
EF
e1
e2
A
RF
Gx
Gy
e1
e2
B C D E FA
RF
Gx
Gy
e1
e2
RF
Gx
Gy
e1
e2
B C D E F
Echo Planar Imaging (EPI)
ky
T2-relaxation
50 55 60 65 70 75 800
5
10
15
20
25
30
35
40
P ixe l pos ition (y)
Re
l. I
nte
ns
ity
ky
T2-relaxation
ky
T2-relaxation
50 55 60 65 70 75 800
5
10
15
20
25
30
35
40
P ixe l pos ition (y)
Re
l. I
nte
ns
ity
∫−
−=
2/
2/
max
max
)exp()()(
k
k
yyy dkjkkPPS
Wyj
WTyS
/21
/1*)2,(
+=
EPI – echo modulation:
Exp T2*-decay =>Lorentzian kernel
W=δy.2/π (ETL . ES/T2*)
Δy= pixel dim in phase enc-dir
−−
−−−⋅=
'
/|,|exp
/),(exp)()(
2
min
2
min
T
vkk
T
vkkTEkrectkd
yyyyyy
yy
EPI – ’damping factor’ in k-space:
Vy = average ’k-space’ speed in y-direction, which is
proportional to SENSE reduction (r-) factor
1/T2* = 1/T2 + 1/T2’1/T2’= ’inhomogeneity induced’
r=1 r=2.4
SE-EPI (3 T)
Jaermann et al MRM (2006)