Statistical Parametric MappingStatistical Parametric Mapping

Lecture 3 - Chapter 5

Hardware for functional MRI

Textbook: Functional MRI an introduction to methods, Peter Jezzard, Paul Matthews, and Stephen Smith

Many thanks to those that share their MRI slides online

The Magnetic Field

• Ferrous Bar Similar to Bar Magnet– Torque – align– Force – toward poles

N

S

N

S

MostlyMostlyTorqueTorque

MostlyMostlyForceForce

Nature of Forces Around Magnet

• Ferromagnetic materials mostly• Depend on shape of object (longer is worse)

• Increase rapidly with approach to magnet (depends on B0 spatial gradient)

• Increase approximately with square of B0 (3T vs 1.5T)

• Depends on type of magnet (open, self-shielded, etc.) (depends on B0 spatial gradient)

fMRI Basic Requirements

• Rapid Imaging– fast high-strength gradients– wide bandwidth transceiver

• Stable System– systematic drift small– noise small

• High Signal Levels– high field strength magnet– RF coil design

Relative to physiological noise

EPI T2* demands

Magnetic Susceptibility

Net B = B0 + B

B is proportional to both field strength (H) and susceptibility (). In air B = 0.

• Macroscopic changes of B induced at different locations result in spatial gradients in B that can be significant for EPI.– For many parts of the brain the macroscopic susceptibility gradient is

small so Larmor frequencies are similar.– For areas where the macroscopic susceptibility gradient is large (e.g.

near tissue air interfaces) Larmor frequencies of nearby voxels also changes greatly.

• Microscopic changes in susceptibility due to BOLD effect can be masked when near areas in brain with large changes in macroscopic susceptibility.

EPI style BOLD fMRI -advantages and disadvantages -

• Fast– Resolve hemodynamic changes, whole head coverage

in 3 seconds or less.– Freeze subject motion (k-space encode of slice in

<50ms).– Encodes full k-space image without RF signal reset

compared to non-EPI imaging (phase errors accumulate).

• Susceptibility weighted– Want good signal from microscopic dephasing due to

BOLD induced susceptibility.– Interference from macroscopic dephasing due to large

extent changes in susceptibility.

Problems With Macroscopic Susceptibility

Gradients

Signal Dropout...

Distortions…

All susceptibility effects increase with Bo!!

BOLD is microscopic susceptibility

Wald, Toronto 2005

Image Encoding for EPI

All lines in one shot…

• Fast (high BW ) in kx.• Slow (low BW) in ky.• No “reset by RF”, so phase

errors accumulate.• Fast (~10 slices per second) for

~2 mm res.• Physiological fluctuations

modulate overall intensity• Readouts alternating polarity.• All k-space NOT treated equally.

dt=0.005msdt=0.5ms

ky

kx

Wald, Toronto 2005

Temporal Sampling is Asymmetric in EPI100x longer in phase direction

• k-space errors due to susceptibility are small in kx direction because of short time sampling intervals.

• but can be significant in ky encode direction (100x longer here).

kx

ky

Wald, Toronto 2005frequency mapfrequency map

=

dephasing leads to signal loss

Note frequency gradient from point 1 to point 2

Image encoding strategies: EPI

Gx

exp(-t/T2*)

t

T2* filtering across k-space increases point-spread function.

• T2* shortens as B0 increases • Limit total readout time to 2T2*

• increase readout gradient• receiver BW increases

All k-space not treated equally:

Wald, Toronto 2005

EPI and Spiral Scanning of k-space

kx

ky

Gx

Gy

kx

ky

Gx

Gy

Wald, Toronto 2005

Interpolated to regular kx and ky spacing.Interpolated to regular kx and ky spacing.

EPIEPI SpiralSpiral

Gx and Gy 90 degrees phase difference for sprial

EPI Spirals

Susceptibility: distortion, blurring,dephasing dephasing

Eddy currents: ghosts blurring

k = 0 is sampled: 1/2 through beginning

Corners of kspace: yes no

Gradient demands: very high pretty high

Wald, Toronto 2005

1 2 3 4 5 6 7 8 9 100

B0, Tesla

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

norm

aliz

ed S

NR

Figure 5.3 from textbook.

Normalized SNR vs. Magnetic Strength

• TAD - total readout time• Time fore single Kx (SE,GRE) TAD << T2*• Time for full K-space (EPI) TAD ~ T2*• TAD intermediate for others (FSE, TSE)

Figure 5.4 from textbook.

SN

RTotal SNR vs Thermal SNR

Signal - SThermal and system noise - 0

Physiological sources - p Total noise -

€

= 02 + σ p

2

SNR =S

σ=

S

σ 02 + σ p

2=

S /σ 0

σ 02 + σ p

2 /σ 0

20

2

000

111 20

22

20

2SNR

SNRSNRSNRSNR

Sp λ

λ

+=

+=

+=

1.

2. 3.

Data from 1.5T (triangles) and 3.0T systems (squares)

Physiological

Figure 5.1b from textbook.

Exciter

Synthesizer

XMTR

T/Rswitch

RFCoil

PreampRCVR

A/P RAMHost

Pulseprogrammer Synthesizer, A/P

XMTR, RCVR, T/R

Shimdriver

Shim coils

Gradient coils

AmpsGx, Gy, Gz

Network

Schematic of MRI System

A/P - Array ProcessorRF, Shim, Gradient Coils inside magnetAll but Host, RAM, and A/P in equipment room

Same or different transmit and receive coil.

Figure 5.7 from textbook.

SN

R

0 50 100 150 200 2500

200

400

600

800

1000

1200

1400Surface coil/head coil comparison

1

2

3

4

17 cm spherical phantom

distance, mm

b

SNR

(1) two surface coils on opposite sides in phase.(2) two surface coils out of phase.(3) single surface coil on right side. (largest SNR)(4) head coil. (most uniform SNR)

RF Coil Uniformity and SNR

(1) (2)

(3) (4)

B1 directions indicated by color arrows.

Surface CoilsSurface Coils

Figure 5.8 from textbook.

Z

(a)Best

(b) Worst

(c)Acceptable

Surface Coil Orientations

Surface coils are like loops for detecting B1 which is precessing about B0 which is parallel to the z-axis

• Best orientation is with plane of coil perpendicular to B0 which for the brain in normal orientation leads to following as best sites

• Left or right side• Anterior of posterior

B0

Tissue Heating During RF Transmit

• Concerns are total body and localized heating• Not practical to monitor increase in temperature except in

phantoms• Specific Absorption Rate (SAR) used to estimate

temperature increase

• 1 SAR = 1 W/kg• 1 SAR would increase temperature of an insulated slab by ~

1 C/hr• SAR also used in monitoring RF for cell phones

Scanner Software Estimates SAR

• Runs a calibration routine• Determines energy for RF pulses• Adds up energy from all RF pulses per TR and divides by TR• Divides by tissue weight to get total body or regional SAR

– Requires height and weight for algorithm

• If limits are exceeded operator must alter pulse sequence

RF - FDA Limits

• Integrated SAR limits– Head SAR = 60 W-min/kg– Trunk SAR = 120 W-min/kg– Extremeties SAR = 180 W-min/kg

• SAR rates– Head (38° C) SAR=3.2 W/kg– Trunk (39° C) SAR =8 W/kg– Extremities (40° C) SAR =12 W/kg

• Other– Infants, pregnancy, cardiocirculatory or cerebral vascular impairment

(1.5 W/kg)

SAR Pulse Sequence Impact

• Minimal for EPI acquisition (1-2 RF pulses per plane)• Higher for 3D anatomical scan GRE (1 RF pulse per kx

reradout) and short TRs.• High for T1W spin echo (one 90º and one 180º RF pulse

per kx line) with slice geometry same as GRE• Within pulse sequence effects

– Increasing TR without increasing # of RF pulses reduces SAR– Reducing number of slices per TR (in multislice SE)– Partial Fourier imaging reduces number of phase encodes with

RF for each (in multislice SE)

Figure 5.5

Transverse (Gx, Gy)

Z

X or Y

Z

X or Y

Longitudinal (Gz)

a b

Gx, Gy, Gz gradient coils

Gx, Gy, Gz shield coils

Z

Figure 5.6 from textbook.

Need strong gradients and shortened readout time to keep TAD in range.

Current and Gradient Pulse ShapeCurrent and Gradient Pulse Shape

a. gradient current supplied (short rise time induces eddy currents)a. gradient current supplied (short rise time induces eddy currents) b. eddy currents oppose changing field w/o compensationb. eddy currents oppose changing field w/o compensation c. gradient current supplied with eddy current compensationc. gradient current supplied with eddy current compensation d. potential field vs time with eddy current compensationd. potential field vs time with eddy current compensation

a

d

c

b

Jerry Allison.

dB/dt Effect (more eddy currents)

Peripheral Nerve Stimulation• dB/dt --> dE/dt• dt is gradient

ramp time• dB/dt largest

near ends of gradient coils

• spatial gradient of dE/dt important

dB

dt

dB/dt / E-Field Characteristics of Stimulation

• Not dependent on B0

• Gradients - 40mT/m (larger Bmax for longer coil)• Gradient Coil Differences - strength (increases dB)

and length (head vs. body determines site)• Rise Time - shorter rise time means shorter dt and

therefore larger dB/dt• Other

– Disruption of nearby medical electronic devices– Subject Instructions

• Don’t clasp hands - closed circuit, lower threshold• Report tingling, muscle twitching, painful sensations

Acoustic Noise Levels

Tomoyuki et al, Toshiba

front rowR&R band

• Earplugs & Headphones– Noise Reduction

Rating – 25-30 dB– Combined 5 dB

more

Ouch

Acoustic Noise

• Lorentz forces acting on gradient coils • Forces & gradient noise level increases with both B0 and

gradient strength• Levels for EPI fMRI

– Peak 130 dB @ 3T, 110 dB @ 1.5T– Average 90-117 dB(A)

• Frequency content varies by sequence– EPI higher average frequency (more read and phase gradients/time)– 3D GRE probably next (short TR)– Spin Echo (depends on TE and TR slices per TR, etc.)

Acoustic Noise

• Lorentz forces acting on gradient coils • Gradient noise level increases with both B0 and gradient

strength• Levels for EPI fMRI

– Peak 130 dB @ 3T, 110 dB @ 1.5T– Average 90-117 dB(A)

• Frequency content varies by sequence– EPI higher average frequency (more read and phase gradients/time)– 3D GRE probably next (short TR)– Spin Echo (depends on TE and TR slices per TR, etc.)

• Experimental Designs– Reduce intra- acquisition noise– Reduce inter-acquisition noise

• Reduce Noise at source– Hardware changes– Gradient shaping

• Passive and Active Noise Reduction– Earplugs, mufflers– Noise reducing headphones

Gradient Noise Management for fMRI

Covered in later lectures.

Active Noise CancellationHeadphones

• Amplitude of sound transmitted to the ear by bone conduction is frequency dependent and maximal at ~2 kHz.

• Active noise cancellation systems may be more useful for 1.5T and 2T systems that produce sounds below 1 kHz.

• Some 3T scanners produce strong sounds in the 1.5-2.5 kHz frequency range.

Additional Equipment

Larissa Stanberry, U of WashingtonLarissa Stanberry, U of Washington

Video Projection Approaches

LCD Projector, Mirror, & Screen

Mirror on RF coil & Screen

Stimulus Presentation / Monitoring

Additional Equipment

• Software• Time-Line

– Control Stimulus

– Monitor Response

– Synchronize timing with MRI

E-PrimeE-Prime

Additional Equipment

• Hardware– Stimulation

• Visual

• Motor

• Auditory

– Response• Visual

• Motor

• Auditory

fMRI Personnel• Patient or Volunteer Support

– Family– Nurse, physician

• MRI Operation– Board Certified Tech

• Research Group– PI & collaborators

• Associated Equipment Tech– Stimulus presentation, monitoring, etc.

• Analysis– PI– Post doc, research assistant, etc.

I know this is not following the theme of this chapter, but important.

fMRI Study Time

• New Design

• Scanning– Setup– Scans– Take down

• Preprocessing

• Statistical Analysis

1-1.5 hr/subject

4+ hr (one instance)

variable

<2 hr/ subject

15-20 min

45 min to 1 hr

15 min

fMRI Study – Raw Data

• Localizer image < 1 MByte

• Anatomy image– Same resolution (2562 x 25) > 3 MByte

– 3D high resolution (2563) > 30 MByte

• Event Related fMRI study– 20 slices/image x 15 images/event x 20 repetitions

x 128x128 images ~200 mByte

– Reorganizing data into volumes indexed by time ~200 mByte

fMRI Study – All Data

• Raw Data ~200 mBytes

• Motion Correction ~180 mBytes

• Other Corrections ~180 mBytes each possibly

• Spatial Normalization ~ 30 mBytes

• Statistical Analysis• Statistical Parametric Image (128x128x20) < 1 MByte

• Statistical Parametric Map (2x SPI) > 1 MByte

Total Data per subject can be 0.5-1.0 gBytesTotal Data per subject can be 0.5-1.0 gBytes

Statistical Parametric MappingStatistical Parametric Mapping

Lecture 3 - Chapter 5

Hardware for functional MRI

Textbook: Functional MRI an introduction to methods, Peter Jezzard, Paul Matthews, and Stephen Smith

Many thanks to those that share their MRI slides online