scan acceleration with rapid gradient-echo · 1 of 214 scan acceleration with rapid gradient-echo...

Post on 23-Mar-2020

1 Views

Category:

Documents

0 Downloads

Preview:

Click to see full reader

TRANSCRIPT

1 of 214

Scan Acceleration with

Rapid Gradient-Echo

Hsiao-Wen Chung (鍾孝文), Ph.D., Professor

Dept. Electrical Engineering, National Taiwan Univ.

Dept. Radiology, Tri-Service General Hospital

2 of 214

The Need for Faster Scan

• Patient comfort, motion artifacts,

efficiency, more information …

• EPI ? You know the difficulty now

• But there are a lot more ways to

accelerate the scanning

3 of 214

Back to the Old Formula

• Scan time for single slice

= TR x (# phase encoding) x NEX

• Reduce phase encoding

– A little faster, trade in resolution

• Reduce NEX (1 or 0.5 at most)

4 of 214

MRI Scan Time (1990 ?)

• Spin-echo : 256x256, 2 NEX

– PD, T2 : 16 min (TR 2000)

– T1 : 5 min (TR 600)

• Note : somewhat exaggerated

5 of 214

Short Scan Time ?

• Reduce TR (2000 50 msec ?)

– 40 times faster ?

– 256x256, 1 NEX : 13 sec

• Sounds like an efficient way ?

6 of 214

Short Scan Time ?

• Reduce TR

– Increased T1-weighting

– Reduced SNR

7 of 214

Effects of Reduced TR on T1 Contrast

TR

Sig

nal In

ten

sit

y

Signal at this TR

Substantial reduction in TR leads to SNR loss

8 of 214

How Low Is The SNR ?

• TR = T1 :

– ~ 63% of thermal equilibrium

• TR = 0.1 T1 :

– ~ 9.5% of thermal equilibrium

9 of 214

Compensating SNR ?

• SNR loss due to slow T1 recovery

– CSF T1 = 0.7 ~ 4.0 sec

• Can magnetization recover from

nonzero (positive) values ?

• Can we retain Mz after RF pulsing ?

10 of 214

Small Flip Angle RF Excitation

Mxy for data acquisition, some Mz for next excitation

z'

y'

x'

Bo Bo z'

y'

x'

11 of 214

The FLASH Technique

• Reducing TR without sacrificing too

much SNR

• Achievable by lowering the flip angle

– via B1 amplitude adjustment

• Fast Low-Angle SHot (FLASH)

– Haase et al., 1985

12 of 214

SNR Comparison

• TR = T1, a = 900 : ~ 63% of Mo

• TR = 0.1 T1 :

– a = 900 : ~ 9.5% of Mo

– a = 250 : ~ 22% of Mo

13 of 214

Example

• CSF has T1 ~ 700 msec

– TR lowered to ~ 70 msec

– Slightly lowered quality

• Scan time ~ 18 sec allows

breath-hold exams

14 of 214

Here Comes A Question

• Spin-echo no longer useable !

• Imaging has to be done with gradient-echo

– Bo inhomogeneity is going to affect

image quality

– But can also become another source of

diagnostic information

15 of 214

Effects of 1800 Refocusing Pulse

Mz becomes negative (recovery takes even longer) !

z'

y'

x'

Bo Bo

z'

y'

x'

16 of 214

Gradient-Echo Properties

• No refocusing function from 1800 pulse

• Image affected by Bo inhomogeneity

– T2* decaying (not T2)

– Instrumentation, air-tissue interface …

– Hemorrhage, hematoma, bone …

17 of 214

Effects of Bo inhomogeneity

Non-uniform Bo in a voxel short T2*

Image voxel :

18 of 214

Effects of TE

TE = 9 msec TE = 18 msec

19 of 214

Gradient-echo Is Worse ?

• Image quality for gradient-echo is often

harder to control than for spin-echo

• But that does not mean “bad”

• Proper usage gives useful information that

cannot be provided by spin-echo

20 of 214

Application Examples

• Hemorrhage (Iron in blood)

• Brain perfusion (Gd-based agent)

• Blood oxygenation (deoxy-Hb)

• Brain fMRI (BOLD contrast)

21 of 214

T2* Signal Loss in Hemorrhage

T1 PD T2 GrE

22 of 214

T2* Signal Loss & Blood Oxygenation

Normal air Pure oxygen

23 of 214

Brain Oxygenation & Brain Function

24 of 214

That Looks Good Then !

• Short TR faster scan

• Small flip angle not much SNR loss

• Gradient-echo more information

• Worth some more exploration !

25 of 214

Effects of Flip Angle

• Small flip angle, partial flip

angle … (< 90)

• How small should it be ?

• 100 ? 300 ? 700 ? Arbitrary ?

26 of 214

Flip Angle Contrast

• Short TR T1WI, long TR T2WI …

– Only true for 900 excitation

• TR already short in gradient-echo

• No longer use TR to alter T1 contrast

27 of 214

How to Vary T1 Contrast?

• Magnetization vector goes into a steady

state after several RF pulsing

• Image intensity mainly determined by this

steady state behavior

• Steady state: T1 recovery for Mz in one TR

= Mz reduction due to RF pulsing

28 of 214

Steady State with Many RF Pulses

Assuming no residual Mxy at the end of TR

Bo Bo

z'

y' x'

29 of 214

The Formula Is Actually

• Signal proportional to

• a : flip angle

(1 - e -TR/T1) sin a

1 - e -TR/T1 cos a e -TE/T2*

30 of 214

Simple Rule for PD or T1 Contrast

T1WI PDWI

z'

y'

x'

Bo Bo z'

y'

x'

recover from 0

little room for

recovery

31 of 214

Control of T1 Contrast

• Large flip angle (~ 900)

– Similar to short-TR images (T1)

• Small flip angle (20 ~ 400)

– Reduced T1 weighting (PD)

32 of 214

Flip Angle = 100

Proton-density-weighted image

z'

y'

x'

33 of 214

Flip Angle = 200

z'

y'

x'

34 of 214

Flip Angle = 300

z'

y'

x'

35 of 214

Flip Angle = 400

z'

y'

x'

36 of 214

Flip Angle = 500

T1-weighted image

z'

y'

x'

37 of 214

Flip Angle = 600

z'

y'

x'

38 of 214

Flip Angle = 700

Strong T1-weighted image

z'

y'

x'

39 of 214

Comparison of PD & T1 Contrast

100 300 500

40 of 214

Control of T2 Contrast

• Still use TE (< TR)

• Actually T2* in gradient-echo

– TE does not have to be too long

– T2* contrast very similar to T2WI

(other than Bo inhomogeneity)

41 of 214

T2(*) Weighting

Decay of transverse magnetization

Bo z'

y'

x'

42 of 214

Using TE to Control T2(*) Contrast

TE = 10 TE = 30 TE = 50

43 of 214

Myth of Speeding Scan

• Is the examination time really

shortened ?

44 of 214

Expansion of a Pulse Sequence …

TR >> TE : hardware mostly idle

Gp

B1 t

t

...

...

Gs t ...

Gr t ...

45 of 214

Add Different Slices

Making best use of the idle time

Gp

B1 t

t

...

...

Gs t ...

Gr t ...

46 of 214

Add Even More Slices

Multi-slice imaging (scan time not lengthened)

Gp

B1 t

t

...

...

Gs t ...

Gr t ...

47 of 214

Myth of Speeding Scan

• As TR shortens, the number of slices

becomes smaller in a TR

• Multiple slice coverage repeat the

scan several times !

• Total exam time likely unchanged

totally useless ??

48 of 214

Pros of Speeding Scan

• Faster single-slice scan

– Less motion influences

• 3D becomes possible !

• 2D examination time is not

necessarily shortened

49 of 214

Scan Time Advantages

6.4 seconds 3.8 seconds

2.5 seconds 1.5 seconds

50 of 214

Speeding Even Further ?

• TR shortened to ~10 msec :

– Flip angle reduced to ~100

– 2-sec scan time !!

– High-quality MR images for

uncooperative patients ?

51 of 214

RF Pulsing with Very Small Flip Angle

Very small TR and flip angle : PDWI

z'

y'

x'

Bo Bo z'

y'

x'

52 of 214

Clinical Restrictions

• TR ~ 10 msec :

– PDWI (seldom used clinically)

– TE < TR

– Then T2 contrast ... ?

53 of 214

Gradient-echo with Very Short TR

TE < TR

Gp

B1 t

t

...

...

Gs t ...

Gr t ...

TR

TE

54 of 214

If You Study Too Hard ...

• CE-FAST, PSIF, SSFP-echo ...

– TE can be larger than TR

• Complicated principles, questionable

applications out of scope !

55 of 214

Clinical Restrictions

• TR ~ 10 msec :

– PDWI (seldom used clinically)

– TE < TR

– Then T2 contrast ... ?

56 of 214

Back to Imaging Principle

• RF with very small flip angle

• T2 & T1 relaxation

• Next RF pulsing

• Repeat many times

57 of 214

RF Pulsing with Very Small Flip Angle

Magnetization almost the same before/after RF

z'

y'

x'

Bo Bo z'

y'

x'

58 of 214

When TR Is Very Short

• RF with very small flip angle

• T2 & T1 relaxation not obvious

• Next RF pulsing

• Repeat many times

• Contrast determined by initial Mo

59 of 214

No Obvious Relaxation ?

• Can be useful !

• Use even smaller flip angle and TR

• Relaxation plays a minor role

• Manipulate the initial magnetization

vector to change contrast !

60 of 214

How to Manipulate “M” ?

• RF pulsing, of course !

• 900 + 1800 + (-900) :

– T2-weighted magnetization !

61 of 214

900-1800-(-900)

Spin echo principle

B1

900 1800

-900

z'

y' x'

62 of 214

900-1800-(-900)

Mz becomes T2-dependent !

B1

900 1800

-900

z'

y'

x'

Short T2

Long T2

63 of 214

Hints for the Processes

• Very small flip angle + very short TR

– T1/T2 do not affect signal intensity

– Contrast determined by initial Mz

• 900 + 1800 + (-900) :

– T2-weighted Mz !

64 of 214

Combine These Two …

Preparation Fast acquisition

Gp

B1 t

t

...

...

Gs t ...

Gr t ...

TR 900 1800 -900

65 of 214

Magnetization Preparation

Preparation Fast acquisition

B1 t ...

TR 900 1800 -900

TE ~ 200 msec TR x 256 ~ 2 sec

Total scan time ~ 2-3 sec

66 of 214

Magnetization Preparation Images (T2)

PD (no prep) T2 (with prep)

67 of 214

Names for The Technique

• Magnetization preparation

• Turbo-FLASH, MP-RAGE (Siemens)

• Driven-equilibrium fast SPGR (GE) ...

68 of 214

Properties

• TR often very short (< 20 msec)

• Flip angle often very small (5~200)

• SNR often low (system-dependent)

• Contrast determined by the

“magnetization preparation” part

69 of 214

The Reason for Low SNR

Not much Mxy available for sampling

z'

y'

x'

Bo Bo z'

y'

x'

70 of 214

Expand the Applications !

• Different preparation modules

• Different acquisition modules

• To form many combinations

71 of 214

Preparation Modules

• STIR/FLAIR (inversion recovery)

• Fat-Sat (off-reson pulse)

• Diffusion (RF + gradient)

• MTC (bipolar pulses) ...

72 of 214

Inversion Recovery Preparation

T1-related preparation or suppression

B1

1800

z'

y' x'

TI

z'

y' x'

short T1

long T1

73 of 214

STIR/FLAIR Turbo-FLASH

B1

1800 TI ~ 130 msec

B1

1800 TI ~ 2000 msec

fat relaxes to 0

CSF relaxes to 0

74 of 214

Fat-Sat Preparation

B1

900 fat only

strong gradient for

spoiling

Gs

Gp

Gr

75 of 214

Other Preparation Schemes

Some will be mentioned in the future

B1

900

Gs

1800

-900

B1

a0 -a0 a0 -a0 a0 -a0 a0 -a0

diffusion

magnetization

transfer

76 of 214

Acquisition Modules

• FLASH, GRASS, SPGR, ...

• EPI (echo-planar imaging)

• FSE (TurboSE)

• Conventional spin-echo !

77 of 214

FLASH (short TR)

Continual RF excitation

Gp

B1 t

t

...

...

Gs t ...

Gr t ...

TR

78 of 214

Echo Planar Imaging

RF

Gs

Gp

Gr

t

t

t

t

79 of 214

Fast Spin-Echo (Turbo Spin-Echo)

RF

Gs

Gp

Gr

t

t

t

t

80 of 214

Preparation + Acquisition

FLAIR Fat-Sat EPI

Gp

B1

Gs

Gr

1800

TI ~ 2000

900

81 of 214

FLAIR Fat-sat EPI

From Picker (Marconi Philips) brochure

Picker Vista

1800 + 2000 msec

TE = 120 msec

256x160

82 of 214

Note (1)

• Short TR gradient-echo actually has

very complicated contrast behavior

• Greatly simplified in this course

• Complex parts saved for the future

83 of 214

Note (2)

• TR > T2 ? TR < T2 ?

• Steady-state and non-steady-state

imaging families

– Approaching steady state

– Destroying steady state (spoiler)

84 of 214

The Fast Spin-echo

Imaging Sequence

Hsiao-Wen Chung (鍾孝文), Ph.D., Professor

Dept. Electrical Engineering, National Taiwan Univ.

Dept. Radiology, Tri-Service General Hospital

85 of 214

Fast (Turbo) Spin-echo Sequence

Every echo forms one k-space line

RF

Gs

Gp

Gr

t

t

t

t

echo 1 echo 2 echo 3 ...

86 of 214

Review: Accelerate Scan?

• Example : EPI

– Fill in the entire k-space after

one single RF excitation

87 of 214

Echo Planar Imaging (EPI)

RF

Gz

Gy

Gx

t

t

t

t

kx

ky

88 of 214

From EPI to FSE

• EPI : series of gradient echoes

– with proper encoding gradient

• FSE : series of spin echoes

– with proper encodign gradient

89 of 214

Echo Planar Imaging (EPI)

RF

Gz

Gy

Gx

t

t

t

t

kx

ky

90 of 214

Fast Spin-Echo (FSE)

RF

Gz

Gy

Gx

t

t

t

t

kx

ky

TR

TR

91 of 214

Also Similar to Spin-Echo

• Spin-echo has the multi-echo option

– 900-1800-echo-1800-echo …

• Multi-echo : forms many images

• FSE : All echoes used in one image

92 of 214

Multi-echo Sequence

Every echo belongs to a unique image

RF

Gs

Gp

Gr

t

t

t

t

image 1 image 2 image 3 ...

93 of 214

Fast Spin-echo Sequence

All echoes belong to the same image

RF

Gs

Gp

Gr

t

t

t

t

echo 1 echo 2 echo 3 ...

94 of 214

What You Can Infer

• Fast spin-echo sequence can be

easily modified from multi-echo

• FSE image behavior should be

similar to traditional spin-echo

95 of 214

Comparison between FSE T2 & SE T2

SE (TE = 100) FSE (TE = 100)

96 of 214

Also Similar to Spin-Echo

• Multiple k-space lines obtained with

every single RF excitation

– Just with several 1800 pulses

• Single-slice scan must be much

faster than spin-echo

97 of 214

20-sec Scan for the Eye

No motion artifacts visible

GE 1.5 Tesla

Fast Spin-echo

ETL = 12

TR = 2000

Scan time = 20 sec

98 of 214

Let’s Name It Then

• Turbo spin-echo (Siemens)

• Fast spin-echo (GE & others)

• RARE (Bruker)

– Rapid acquisition with relaxation

enhancement

99 of 214

Why Is FSE Important ?

• Spin-echo : traditional MRI standard

• FSE similar to SE

• Much faster scan

– TR = 2000 : 7 min to 1 min

100 of 214

FSE Similar to SE (256x128)

SE (6 min) FSE (48 sec)

101 of 214

Acceleration Achieved

• 8 echoes (e.g.) with each 900

• 256x256 : 32xTR only

• 8 times faster than spin-echo

• Echo train length (ETL) = 8

102 of 214

Multi-shot FSE

• Scan time = TR x (phase#) / ETL

• The larger ETL, the faster single-

slice scan

103 of 214

How about Contrast ?

• Echoes have different TE !

• What determines T2 contrast ?

• Effective TE

104 of 214

Multi-shot FSE Sequence

The k-space lines have different TE ??

RF

Gz

Gy

Gx

t

t

t

t

Signal t

kx

ky

TR

TR

105 of 214

Don’t Forget about k-space …

Contrast mainly determined by central k-space

kx

ky

boundary

boundary

contrast

106 of 214

A 256x256 Image Is Composed of …

Central k: contrast Outer k : boundary

107 of 214

TE in FSE

• Central k-space determines the

image contrast

• Data passing central k-space

dominate the contrast

– despite of different TEs

108 of 214

TE & Phase Encoding

• Data location in k-space

controlled by phase encoding

• Phase encoding order

determines TEeff

109 of 214

k-space Filling Pattern in FSE

Early echo placed at central k-space: PDWI

RF

Gz

Gy

Gx

t

t

t

t

Signal t

kx

ky

110 of 214

k-space Filling Pattern in FSE

Late echo placed at central k-space: T2WI

RF

Gz

Gy

Gx

t

t

t

t

Signal t

kx

ky

111 of 214

Expand: Dual-Contrast

• FSE is an expanded version of multi-

echo spin-echo …

• Dual echo naturally feasible in FSE

• T2 weighting also determined by

TEeff

112 of 214

Dual-contrast FSE Sequence

RF

Gs

Gp

Gr

t

t

t

t

echo 1 echo 2 echo 3 ...

image 1 image 2

113 of 214

Or Even ...

Data sharing

Contrast

Early echo : PD

Late echo : T2

kx

ky

boundary

(shared)

boundary

(shared)

114 of 214

Data Sharing FSE Sequence

RF

Gs

Gp

Gr

t

t

t

t

echo 1 echo 2 echo 3 ...

image 1 image 2

115 of 214

Data Sharing in Dual-Echo FSE

TEeff = 17 msec TEeff = 85 msec

116 of 214

Data Sharing

• Only central k-space acquired

multiple times with different TE

– For different T2 weightings

• Outer k-space acquired only once

• Dual contrast with < 2 time penalty

117 of 214

Move Further: Single-shot

• Entire acquisition + wasted time

< 1~2 T2 (100 ms to sec range)

• 256x256 : an echo every 4 msec

– Echo spacing (ESP)

118 of 214

Multi-shot FSE Sequence

Scan time = TR x (phase #) / ETL

RF

Gz

Gy

Gx

t

t

t

t

TR

ESP

ETL = 3

119 of 214

Single-shot FSE Sequence

No TR (or TR is infinite)

RF

Gz

Gy

Gx

t

t

t

t

ESP

ETL = # of phase encoding

120 of 214

Certainly Possible But …

• ESP has ~4 ms lower limit

• ETL ~ 256 to yield 1-2 sec scan

• Most signals decay due to T2

relaxation

121 of 214

Single-shot FSE Sequence

Very late echoes show no signals at all

RF

Gz

Gy

Gx

t

t

t

t

Signal t

kx

ky

122 of 214

ESP Can’t Be To Short !

• Specific Absorption Rate (SAR)

• RF power proportional to (flip angle)2

– 1800 power: 4x of 900, 36x of 300 !

• RF power deposition causes an

increase of local body temperature

123 of 214

Single-shot FSE Sequence

So many high-power RF pulses !

RF

Gz

Gy

Gx

t

t

t

t

ESP

ETL = # of phase encoding

124 of 214

Single-shot FSE Usage

• You want only long T2 tissues

– Myelogram, MRCP

• Motion so severe that scan time

becomes the dominant factor

– Fetal imaging, GI imaging

125 of 214

Only Long-T2 Tissues Have Signals

CSF in spinal cord : long T2 tissue

126 of 214

Myelogram (Strongly T2-weighted FSE)

Original slices (heavy T2 images) MIP

127 of 214

FSE MRCP (Same Principle)

Original slices MIP MRCP

128 of 214

1-sec Fetal Scan

No artifacts from fetal motion

Siemens 1.5 Tesla

HASTE

ETL = 128

256x240

Scan time = 1 sec

22 weeks gestation

129 of 214

… Including My Own Son

Courtesy Cheng-Yu Chen, M.D., Tri-Service General Hospital

5-month photo Future look?

28-week gestation 35-week gestation

130 of 214

One Variation : HASTE

• Half-Fourier acquisition single-shot

TurboSE (Siemens)

• Single-shot fast spin-echo (GE)

• Half Fourier + TSE = ~1s scan

• Reduce 1800 to ~1300 for SAR

131 of 214

Multi-shot FSE Usage

• Almost the new standard for T2

– Much faster than traditional SE

• HASTE best in GI

– Motion and susceptibility artifacts

132 of 214

FSE Advantages

• FSE similar to traditional SE

– Spin-echo already a standard

– FSE widely accepted as well

– No gradient-echo artifacts

133 of 214

Comparison between FSE T2 & SE T2

SE (TE = 100) TSE (TE = 100)

134 of 214

Speed Advantages

• Overcome motion artifacts

• Multiple signal averages for SNR in

reasonable scan time

• Trade SNR for spatial resolution

• Long TR for proton density weighting

135 of 214

Speed Advantages in terms of Motion

SE (ECG gating) FSE (no gating)

136 of 214

Speed Advantage in GI Imaging

4:30 min scan, 512 matrix (readout)

R-L frequency encoding

137 of 214

Resolution Advantage with SNR

High-resolution in reasonable scan time 256x256, 57 sec 512x512, 2:45 min

138 of 214

Long TR Advantage in Nerve Roots

Strong CSF & high resolution for nerve roots

Siemens 1.5 Tesla

Turbo Spin-echo

512 matrix

3 mm slice

Scan time = 7 min

139 of 214

FSE Properties

• Compared with SE at same TE

– Stronger magnetization transfer

contrast

– Weaker diffusion weighting

– Bright fat at long ETL

• No time to explain in this semester

140 of 214

FSE Unique Artifacts

• Point-spread function blurring

– Will be briefly mentioned

• Pseudo edge enhancement

• Ghosts from data discontinuity

• No time to explain either

141 of 214

k-space Filling Pattern in FSE

Early echo placed at central k-space : blurring

RF

Gz

Gy

Gx

t

t

t

t

Signal t

kx

ky

142 of 214

Blurring in FSE with Long ETL

HASTE (176x256) HASTE (128x256)

143 of 214

ETL Comparison in Chest Imaging

ETL 15 (ECG, BH, 14 sec) ETL 85 (0.4 sec)

144 of 214

Parallel MRI with

RF Phased Array Coils

Hsiao-Wen Chung (鍾孝文), Ph.D., Professor

Dept. Electrical Engineering, National Taiwan Univ.

Dept. Radiology, Tri-Service General Hospital

145 of 214

Review : Phased Array

• Surface coil: high SNR with limited

coverage

• Phased array: multi coils with geometric

arrangement to cancel mutual inductance

• Achieve high SNR and wide coverage

simultaneously

146 of 214

Phased Array Coil

147 of 214

Spine Phased Array

148 of 214

Phased Array Image Formation

Signals received and processed separately

Receiver Receiver Receiver Receiver

Computer (reconstruction)

149 of 214

Combine to Form Phased Array Image

Wide FOV for larger coverage

150 of 214

Phased Array Imaging

• Coil elements receive signals

separately

• Send to individual receiver channel

• No other difference at all

– RF pulsing, phase encoding, etc.

151 of 214

What Is Parallel Imaging ?

• Signals in different coils must be different

• If data in different coils show little

redundancy, can some steps be omitted?

• SMASH (1997),SENSE (1999)

152 of 214

Method 1

• Produce various spatial

frequency waveforms in k-space

using the coil profiles

• Multiple k-space lines in one

phase encoding

153 of 214

Review : k-space & MRI

• Each point in the k-space coordinate

– (kx,ky) coordinate : specific waveform

– Signal intensity : relative weighting of

that waveform

• All MRIs are formed by these waveforms

154 of 214

kx

ky

A k-space point represents a waveform

155 of 214

Many waveforms summed to an MRI

Waveforms weighed by signal intensity

+

+

+

+ …

156 of 214

Phased Array Helps in …

• Signals received at various locations

• Adjust weights of signals according

to the coil locations to “form”

different waveforms

– from one single acquisition

157 of 214

Waveform Formation from Coil Profiles

8 elements arranged linearly

Coil arrangement

Equal weights

Form cosine

Form sine

High-freq cosine

High-freq sine

158 of 214

Many Waveforms from One Acquisition

Many k-space lines from one phase encoding

kx

ky

freq encoding

phase encoding

159 of 214

Parallel Imaging

• Multiple k-space lines with one phase

encoding due to separate signal

receiving with phased array coils

• N waveforms N times acceleration

160 of 214

Let’s Name It

• Many “harmonics” formed at once

• SiMultaneous Acquisition of Spatial

Harmonics (SMASH)

– Sodickson 1997

161 of 214

Acceleration Factor

• Theoretically, N coil elements

could form at most N harmonics

• Nothing is perfect in practice

acceleration factor < N

162 of 214

SMASH Phantom Image (1997 MRM)

Usual scan (10 sec) 3 coils (5 sec)

163 of 214

SMASH Body Image (1997 MRM)

Usual scan (22 sec) 4 coils (11 sec)

164 of 214

Philips ACS NT 1.5T

FLASH, 7.0/1.5/300

3D (128px256x20)

6 coils, R = 3

[Gd] = 0.13 mM/Kg

8 sec per 3D frame

SMASH CE-MRA (2000 Radiology)

Shorten breath-hold time or high temporal resolution

165 of 214

SNR in SMASH

• Accelerate from reduced phase encoding

• SNR lowers according to the square root

relationship

• Half scan time SNR lowered to 70%

• Used when reducing motion effects

outweighs SNR loss

166 of 214

SMASH Pitfalls

• Coil size, shape, arrangement

relatively restricted in order to form

perfect sinusoids

• Direction of multi coils often not

used for phase encoding

167 of 214

Waveform Formation from Coil Profiles

8 elements arranged linearly

Coil arrangement

Equal weights

Form cosine

Form sine

High-freq cosine

High-freq sine

168 of 214

Combine to Form Phased Array Image

Head-foot direction is often freq encoding

169 of 214

Even Arrangement OK …

• Sinusoids formed by coil profiles

often non-perfect

• Imperfect reconstruction results

in residual aliasing

170 of 214

Aliasing in SMASH with R = 2

Usual scan (10 sec) 3 coils (5 sec)

171 of 214

Aliasing in SMASH with R = 2

Usual scan (22 sec) 4 coils (11 sec)

172 of 214

SMASH Extensions

• Auto-SMASH

• VD Auto-SMASH

• GRAPPA (Siemens)

• Details omitted

173 of 214

GRAPPA Lung Images

Usual scan 207 ms 150 ms (8-coil array)

HASTE 128x256 GRAPPA 256x256

174 of 214

GRAPPA Liver Images

Usual scan 252 ms 252 ms (8-coil array)

HASTE 128x256 GRAPPA 256x256

175 of 214

Method 2

• Coils have different sensitivity profiles, all

relatively small

• Reduce FOV for less phase encoding

– Accelerated, aliasing occurs

• Compute image according to the different

aliasing patterns

176 of 214

3-coil Example

• Coil sensitivity profiles roughly

occupy 1/3 FOV

• Prescribe a small FOV (~1/3)

• Resolution unchanged

reduced matrix size

177 of 214

Example Using 3 Coils

Phantom & coil locations Aliased images

1 2

3

1

2

3

178 of 214

No Panic with Aliasing

• Aliased image =

signals within FOV + outside FOV

• Signals stronger within coil profile,

weaker outside weighted sum

179 of 214

Aliased Image from Coil 1

Phantom & coil location

1

FOV

aliasing

aliasing

180 of 214

Aliased Image from Coil 1

Coil #1 local intensity

FOV

aliasing

(weaker)

aliasing

(even weaker)

1

181 of 214

Aliased Image from Coil 1

Phantom & coil image

aliasing

(weaker)

aliasing

(even weaker)

1

182 of 214

Aliased Image = Weighted Sum

• An aliased images (D1) =

weighted sum of 3 sub-FOV images

• In the form of D1 = A1 x + B1 y + C1 z

183 of 214

Aliased Image from Coil 2

Coil #2 local intensity image

aliasing

(weaker)

aliasing

(weaker)

2

184 of 214

Aliased Image from Coil 3

Coil #3 local intensity image

aliasing

(weaker)

aliasing

(even weaker)

3

185 of 214

3 Aliased Images from the 3 Coils

Phantom & coil location Aliased images

1 2

3

1

2

3

186 of 214

3 Aliased Images (D)

• D1 = A1 x + B1 y + C1 z

• D2 = A2 x + B2 y + C2 z

• D3 = A3 x + B3 y + C3 z

• Solving the equations (matrix

inversion) gives (x, y, z)

187 of 214

Algebraic Problem Now

• 3 aliased images (D)

– in “D = Ax + By + Cz” form

• A, B, C: known from coil profiles

• Matrix inversion to get (x, y, z)

188 of 214

Don’t Forget …

• Each aliased pixel has one

unique set of D = Ax + By + Cz

equations

• Matrix inversion performed

256x256/3 times (for R = 3)

189 of 214

Summary …

• 3 RF coils to receive signals

• 1/3 FOV prescribed with same resolution

• Scan accelerated 3 times from a reduction

in matrix size (phase encoding)

• Full FOV image can be computed

190 of 214

3 Times Acceleration, Solve Equations

Full FOV image obtained from aliased images

1

2

3

191 of 214

Let’s Name It

• It is about the usage of coil

sensitivity profiles …

• SENSitivity Encoding (SENSE)

– Pruessmann 1999

• Rumor has it that it’s Philips patent

192 of 214

Acceleration Factor

• Theoretically, N coil elements

provide at most N aliased images

– Smallest prescribed FOV is FOV/N

• Like SMASH, < N in reality

193 of 214

SENSE Brain Image (1999 MRM)

Usual scan (170 sec) 2 coils (85 sec)

194 of 214

SENSE Heart Image (Short Axis)

Usual scan (15 beats) 5 coils (5 beats)

195 of 214

SNR in SENSE

• Acceleration thru reduced phase encoding

• SNR lowers according to square root

relationship

• Like SMASH, Used when reducing motion

effects outweighs SNR loss

196 of 214

SENSE Heart Image (Axial)

Usual scan (128x128) 6 coils (R = 3)

197 of 214

SENSE Coil Requirement

• D1 = A1 x + B1 y + C1 z

• D2 = A2 x + B2 y + C2 z

• D3 = A3 x + B3 y + C3 z

• Solvable as long as equations

are linearly independent

198 of 214

SENSE RF Coil Arrangement

Phase direction can be either one

Phase

Phase

RF coil element

199 of 214

SENSE Coil Requirement

• Linearly independent equations

• No need to form perfect sinusoid

• Easier than SMASH

• No many variations like GRAPPA

200 of 214

SENSE Pitfalls

• Incompatible with restricted FOV

• Example : cardiac imaging

• Full FOV contains some aliasing

• Can’t distinguish after mixture

with 1/3 FOV aliasing

201 of 214

Theoretical Comparison

• SMASH :

• Fast computation (Fourier transform)

• Artifact performance better than

SENSE at high acceleration factors

202 of 214

Theoretical Comparison

• SENSE :

• Coil arrangement flexible

• Flexible slice orientation as in MRI

• General artifacts less than SMASH

203 of 214

Practical Comparison

• SENSE & SMASH performance

highly depends on human

resources devoted to R&D

• No major difference when

commercialized

204 of 214

Parallel Imaging Families

• Philips : SENSE

• Siemens : iPAT

– GRAPPA + mSENSE

• General Electric : ASSET

205 of 214

Parallel MRI Advantages

• Phased array coils have long been

commercialized (’94)

• Matrix inversion software simple

• Acceleration is basically pulse-

sequence independent

206 of 214

SENSE Coronary Angiogram

Usual scan (3.0 T) Similar quality at 3x

207 of 214

Speed Advantage of SENSE

Abdominal CE-MRA 512x512 T2W FSE

208 of 214

More Advantages

• Shorten EPI acquisition time

– Less EPI geometric distortion

• Reduce ETL in FSE

– Less blurring in HASTE

209 of 214

SMASH EPI of Brain (Sagittal)

Usual scan Four coils

210 of 214

SENSE DW-EPI of Brain (Axial)

8 coils (R = 4) to reduce distortions (3.0T)

211 of 214

SMASH HASTE of Chest (192x256)

Usual scan Four coils

212 of 214

Trade Scan Time for Resolution (R = 2)

Usual scan

192x256, 450 ms

4 coils

192x256, 225 ms

4 coils

384x256, 450 ms

213 of 214

Parallel MRI Advantages

• Major medical centers will have it

soon after its first introduction

• Taiwan will have it after 3-5 years

at the latest time (a matter of $$)

214 of 214

The Fast Imaging

Techniques

Hsiao-Wen Chung (鍾孝文), Ph.D., Professor

Dept. Electrical Engineering, National Taiwan Univ.

Dept. Radiology, Tri-Service General Hospital

top related