Introduction to

MRI Physics
Slides originally by Karla Miller, FMRIB Centre

Modified by Mark Chiew (mark.chiew@ndcn.ox.ac.uk)

Slides available at:

http://users.fmrib.ox.ac.uk/~mchiew/teaching/
MRI Physics
Monday:
★ Basics of magnetic resonance
★ Image formation
★ Signal statistics (SNR)
★ Functional MRI

Wednesday:
★ Image contrast (T2 and T2*)
★ Spin vs. gradient echo
★ Fast imaging
★ Diffusion MRI
 What are we trying to achieve?
Informed decision making: You need to take responsibility for
the design, implementation & execution of your study
• Protocols need to be tailored to the problem
• Learning some physics will make this less daunting

A common language: You need to be able to talk to experts

• Communicate your needs to physicists/radiographers/techs
• Build an MR vocabulary (terminology/jargon)
• Gain some intuition behind imaging concepts
MRI Physics

★ Basics of magnetic resonance

★ Image formation spin

★ Signal statistics (SNR)

★ Functional MRI

1H
“Spin” spin
N

1H

Almost all sub-atomic particles have “spin”

• All nuclei with odd numbers of protons/neutrons
will have non-zero net spin
 1H, 13C, 23Na, 31P

All hydrogen protons will act like

little magnets

H
H

The abundance of water in the human body

makes this very powerful!
 1H, 13C, 23Na, 31P

Can also do this with phosphorous nuclei

ATP
Spins, or magnetization
(when referred to in bulk)
behave similarly to classic
physical systems

In many ways analogous

to simple oscillators, like
swings or pendulums
1. Excitation
Magnetization can be moved or
rotated by applying “excitation”
magnetic fields (RF)

2. Resonance
Magnetization will “resonate” at
a frequency proportional to
magnetic field strength

3. Relaxation
The oscillations die out, i.e.
magnetisation “relaxes” back to
equilibrium – speed of relaxation
is tissue-dependent!
 The External Magnetic Field (B0)
N
N
spin

S (no magnetic field) S

Normally: protons randomly oriented no net magnetism

External field: protons align slightly net magnetization (M)

Only a few parts-per-million!

 Magnetic resonance
Magnetic: external field (B0) magnetizes sample

This “Larmor Equation” 


ω0 = γB0 defines the resonant 

frequency

Resonance: magnetization has characteristic (resonant)

frequency proportional to external field B0
Coordinate system
z
M
B0
y
x
B0

Direction of main field (B0) defines coordinate system

Longitudinal axis: parallel to B0 (typically z)
Longitudinal magnetisation: Portion of M aligned with B0
Coordinate system
z
M
B0
y
x
B0

Direction of main field (B0) defines coordinate system

Transverse plane: perpendicular to B0 (typically x,y)
Transverse magnetisation: Portion of M perpendicular to B0
The Basic MRI Experiment: 

1. Excitation

B0

courtesy of William Overall

ω0 = γB0
Excitation pulse (yellow) tips magnetisation away from B0
Excitation must occur at the resonant frequency ω0
The Basic MRI Experiment: 

1. Excitation

B0

courtesy of William Overall

In a frame that rotates with B1, magnetisation is simply

“flipped” or “tipped” out of alignment with B0

Hence the term “flip angle” or “tip angle”

The Basic MRI Experiment: 

2. Resonant Precession

B0

courtesy of William Overall

ω0 = γB0
Once excited, magnetisation precesses/oscillates/rotates
at resonance frequency
The Basic MRI Experiment: 

3. Relaxation

B0

courtesy of William Overall

As it precesses, it also “relaxes” back into alignment with B0

Speed of relaxation has time constants: T1, T2, T2*, which
relate to the image contrast
The Basic MRI Experiment: 

3. Relaxation

B0

courtesy of William Overall

T1 : describes speed of recovery along longitudinal (z) axis

T2, T2* : describe speed of signal decay in transverse (x-y) plane
The Basic MRI Experiment: 

4. Signal Detection

B0

courtesy of William Overall

As the magnetization precesses and relaxes

The precession induces a voltage in the receive coils
Coils only detect rotating, transverse magnetisation
 Magnetic fields everywhere…

B0 B1 Gx,Gy
Main magnetic field (B0): always on, static
Excitation RF field (B1): pulsed on & off, 60-300 MHz
Magnetic field gradients (G): pulsed on & off, “static”

MRI scans: carefully timed RF and gradient “pulse sequences”

MRI Physics

★ Basics of magnetic resonance

★ Image formation
★ Signal statistics (SNR)
★ Functional MRI
 Magnetic Field Gradients

G
B0 magnet

Differentiate between signal from different locations

Add a spatially varying magnetic field gradient (G)

• Field varies linearly along one direction
• Gradient field adds to or subtracts from B0
No Gradient
x-Gradient
y-Gradient
Precession

B0

courtesy of William Overall

ω0 = γ(B0+G(x,y,z))
Resonance frequency is proportional to total field:
Static B0 + applied gradients
 Gradients and Resonance
ω0 = γ(B0+G(x,y,z))
Higher
Higher field frequency

B0

Lower field Lower

frequency

Distinguish different spatial locations by assigning different

resonant frequencies to different positions
 Imagine differentiating between instruments
based on their frequency content!

Highest Middle Lowest

frequency frequencies frequency
 Frequency decomposition
Fourier
Transform
Amplitude

time (s)
Cellos
(playing the loudest)

Amplitude
Bass
Violins

Other instruments
(playing quietly)
Inverse
Fourier frequency (Hz=s-1)
Transform
Simple “imaging” experiment (1D)

increasing x
field
 Simple “imaging” experiment (1D)

Signal
time

Fourier transform

“Image”
frequency/
position

Fourier Transform: Gives us the “frequency content” of

our signals
Simple “imaging” experiment (1D)

Signal
time

Fourier transform

“Image”
frequency/
position

This is “frequency encoding”

 Magnetic gradients

Gx,Gy

It’s a bit more complex in more than 1 dimension

Have 3 gradient fields (along x, y, z)
Manipulate the strength & timing independently
Gradients in multiple dimensions
 Magnetic field gradients

B0
Gradients in multiple dimensions
 Combined field gradients

B0
 Spatial frequencies or patterns

At any instant in time, signal is across

space is defined by a specific “pattern” of
the magnetisation phase (orientation),
i.e., its spatial frequency that depends on
the applied gradients

Spatial frequencies:
• sinusoidal pattern over
space instead of time
• extend to multiple
dimensions
Gradients and Spatial Frequency
Strong,
positive
gradient

zero
gradient

Strong,
negative
gradient
Gradients and Spatial Frequency
higher stronger
gradient
resonance magnetic
frequency field

faster
precession

lower
resonance
frequency
slower zero
precession gradient
This is one spatial frequency...
y

2 cycles
along y:
ky=2

signal
signal

x 0 cycles along x:
kx=0
This is another one...
y

0 cycles
along y:
ky=0

signal
signal

x 2 cycles along x:
kx=2
This is another one...
y

2 cycles
along y:
ky=2

signal
signal

x 2 cycles along x:
kx=2
Each of these represents one 2D
pattern or frequency: denote (kx,ky)
(2,0) (0,2) (2,2)

(2,1) (1,2) (4,4)

(4,0) (0,4) (8,1)

“k” values are the number of cycles in each direction

2D “k-space” describes contribution
of each spatial frequency

Patterns determine the

“where” in k-space

x
x
(0,4) ky=0

(2,1)

kx=0
2D “k-space” describes contribution
of each spatial frequency
Sum total signal after application of
these patterns determines the
“value” of each k-space location

x
x
(0,4) ky=0

(2,1)

kx=0
Signal from RF coil

Filling Fourier
k-space transform
ky

kx
“k-space”
y

x
Image
Think of each pattern (k-space
location) as a filter on a camera
(2,0) (0,2) (2,2)

(2,1) (1,2) (4,4)

(4,0) (0,4) (8,1)

Imagine our “camera” can only see
one colour at a time
Imagine our “camera” can only see
one colour at a time (blue filter)
Imagine our “camera” can only see
one colour at a time (red filter)
Imagine our “camera” can only see
one colour at a time (green filter)
Combine the filtered images to
form the final image
Scanner takes a series of measurements with each k-
space “spatial filter” (as many filters as voxels)
The “spatial filters” are applied using gradients

Measurements are then combined using the

Fourier Transform to form image
Scanner takes a series of measurements with each k-
space “spatial filter” (as many filters as voxels)
Higher resolution means “finer” features,
which require “finer” filters

More spatial resolution → more voxels and

filters needed → longer acquisition time
If gradient is on, spatial
frequencies change
over time

ky

kx
Visualize as gradients “move us” through
k-space!
If gradient is on, spatial
frequencies change
over time

ky

kx
Visualize as gradients “move us” through
k-space!
 Signal from RF coil
The trajectory is the ordering
of k-space data acquisition

k-space
trajectory Fourier
transform

“k-space”
y
Trajectory = Path through k-space
or the sequence of spatial filters x
Image
Linescan (2DFT) Acquisition

kx

ky

Acquire one line after each excitation

Linescan (2DFT) Acquisition

kx

ky

Acquire one line after each excitation

Useful for structural images (minimal artifacts)
Echo-planar Imaging (EPI) Acquisition

kx

ky

Acquire all of k-space in a “single shot”

Used for FMRI, diffusion imaging
 Slice Selection

ω0 frequency
Transmit all frequencies
corresponding to desired slice gradient

excited slice
2D Multi-slice Imaging

excited slice

t1
t2
t3
t4
t5
t6
Slices excited and acquired sequentially (separately)
Most scans acquired this way (including FMRI, DTI)
Simultaneous Multi-slice Imaging

z=1+4 z=2+5 z=3+6

t1 t2 t3

}
“Multi-Band” Factor 2
MRI Physics

★ Basics of magnetic resonance

★ Image formation
★ Signal statistics (SNR)
★ Functional MRI
Signal-to-noise ratio (SNR)

high SNR low SNR

Signal (magnitude)
SNR =
σnoise (standard deviation)
Signal-to-noise ratio: describes signal “robustness”
All else being equal, we want to maximize SNR!!
Signal-to-noise ratio (SNR)
Protocol choices affecting SNR...

• RF receive coil & field strength

• Timing: bandwidth, TE & TR
• Voxel volume
• Scan duration (imaging time)
• Anything affecting signal!!!
SNR and acquisition time or averages

σnoise

scan time

Longer acquisition less noise higher SNR

SNR improves with the square root of scan time
i.e., to double SNR you need to scan 4x longer
SNR and voxel volume

? 8x
SNR

Larger voxels have signal from more tissue!

– Signal proportional to voxel volume
– 2x2x2mm has 8x higher SNR than 1x1x1mm!
Averaging to achieve high resolution

? 8x
SNR

Can we recover lost SNR by averaging?

Yes! But requires a 64-fold increase in scan time
(because you only get square root benefit)
Contrast-to-noise ratio (CNR)
MRI Physics

★ Basics of magnetic resonance

★ Image formation
★ Signal statistics (SNR)
★ Functional MRI
 A source of signal loss: dephasing

B0

When spins are “in-phase”, they are all oriented the same way
Over time, the spins within a voxel lose alignment (“dephase”)
Apparent increase in T2 = T2*

signal
Perfect field
(T2 decay)
Imperfect field
with inhomogeneity
(T2* decay)
time

Dephasing causes magnetization vectors to partially

“cancel” each other out
Dephasing results in a lower net signal magnitude
Apparent decrease in T2: called T2* (more on Wednesday)
Deoxyhaemoglobin is the source of FMRI signal
Deoxyhaemoglobin is the source of FMRI signal

O
Fe2 O
Fe2

O
Fe2 O
Fe2

When oxygen is bound to the haemoglobin, it shields the

magnetic effects of iron atoms in the heme groups
Deoxyhaemoglobin is the source of FMRI signal

Fe Fe

Fe Fe

Without oxygen, the iron (Fe) is exposed, causing magnetic

field inhomogeneities due to its strong magnetic properties
Field inhomogeneity leads to T2* change (FMRI signals)
 The BOLD Effect [ Ogawa et al, 1990 ]

imaging voxel

Blood Oxygenation Level Dependent (BOLD) effect

Vessels, depending on orientation and blood oxygen content will
alter their local magnetic fields
BOLD Effect – vessel orientation

Water Water Water

Vessel Vessel Vessel

0º 45º 90º
Vessel parallel Vessel perpendicular
BOLD Effect – vessel size

Strength of Magnetic
B0 direction
Field Inhomogeneity

Water Water Water

Vessel Vessel Vessel

radius = 50 μm radius = 100 μm radius = 150 μm

BOLD Effect – blood oxygenation level

Strength of Magnetic
B0 direction
Field Inhomogeneity

Water Water Water

Vessel Vessel Vessel

Oxygenation = 60% Oxygenation = 30% Oxygenation = 0%

Oxygenation: Y=60%
More Oxygenated Hb
Low inhomogeneity
Longer T2*
Higher signal

BOLD
Contrast

Oxygenation: Y=0%
More de-oxygenated Hb
High inhomogeneity
Shorter T2*
Lower signal
 Vascular Response to Activation

neuron

capillary
HbO2 dHb HbO2 HbO2 HbO2
HbO2 HbO HbO2
dHb2HbO HbO dHb dHb HbO
dHb2
2 2 HbO2
HbO2 HbO2 HbO2
HbO2 dHb HbOdHb2 HbO2
HbO2 HbO HbO2 HbO2 HbO dHb2 HbO2
dHb HbO2
2
HbO2 HbO2

HbO2 = oxyhemoglobin
O2 metabolism dHb dHb = deoxyhemoglobin

blood flow HbO2 [dHb]

blood volume HbO2
 BOLD Contrast
1.0

0.8 optimal
range
O2 use

BOLD signal
0.6 difference
(contrast)
blood
flow [dHb]
0.4
blood [dHb]
volume rest active
0.2
[dHb]

20 40 60 80
Echo time (TE, ms)

Signal increases during activation (less decay)

Signal change for longer delay (TE)
Typically, 1–5% signal change
BOLD signal and field strength (B0)

SNR and BOLD effects can increase with field strength

But image artefacts get worse at higher field strength
3T is currently a good tradeoff of signal vs artefacts
Sources of BOLD Signal
Blood flow

Neuronal Metabolism BOLD

activity [dHb] signal

Blood volume

Indirect measure of activity (via metabolism!)

Subject’s physiological state & pathology can change
neurovascular coupling, muddying interpretation
Hemodynamic response function (HRF)
on

Stimulus
off
timing
time
BOLD
response

Vascular response to activity is delayed & blurred

Described by “hemodynamic response function”

Limits achievable temporal resolution

Must be included in signal model
 What is required of the scanner?

image 1 2 3 … TR

Typical stimulus lasts 1–30 s

Rapid imaging: an image every few seconds
Anatomical images take minutes to acquire!
Acquire “single-shot” images (e.g., EPI)
Typical* FMRI Parameters
* Typical, not fixed!!

Parameter Value Relevant points

TE 1.5T: 60 ms Determines functional

(echo time) 3.0T: 30-40 ms contrast, set ≈T2*
7.0T: 15-25 ms
TR 1–4 s HRF blurring < 1s;
(repeat time) Poor resolution > 4s
Matrix size / 64x64 – 96x96 Limited by distortion,
Resolution 2–3 mm SNR, FOV
Scan 2-15 mins Lower limit: sensitivity
duration Upper limit: compliance
Confounds: Noise

time

Purely random noise Structured noise

(example: “thermal”) (example: “physiological”)

Noise: signal fluctuations leading to less robust detection

with respect to statistical measures
Confounds: Artefacts

Dropout Distortion “Ghosting”

Artefacts: systematic errors that interfere with

interpretability of data/images
 Source of signal dropout

BOLD contrast is based on signal dephasing

BOLD imaging requires longish delay (TE) for contrast
 Dropout is just extreme dephasing

sinus

Dephasing also occurs near air-tissue boundaries

Sensitivity to BOLD means signal loss near air-tissue boundaries
 BOLD Signal Dropout

Short TE Long TE
Dephasing near air-tissue boundaries (e.g., sinuses)
BOLD contrast coupled to signal loss (“black holes”)
Air-tissue effect is often larger than BOLD effect
Dropout is not correctable post-acquisition!
 Image distortion
field offset local warping

Field map EPI

We think frequency maps to spatial location...
So errors in frequency cause spatial mis-localization!
More on Wednesday...
 Final thoughts

Understand how different experimental parameters affect

SNR and image artefacts

Tradeoffs: you can’t get something for nothing, but you do

have options

Get to know an engineer/physicist/radiographer: get help

setting up study protocols, show them your artefacts

Quality assurance: always look at your data, even if you are

running a well-tested protocol
 Questions:

mark.chiew@ndcn.ox.ac.uk