1. Introduction to Functional MRI
MAIN SOURCES:
FMRI Graduate Course (NBIO 381, PSY 362)
Dr. Scott Huettel
Duke-UNC
Brain Imaging & Analysis Center (BIAC)
Jody Culham
Brain and Mind Institute
Department of Psychology
University of Western Ontario
Karla L. Miller
FMRIB Centre
Oxford University

2. The First “Brain Imaging Experiment”
… and probably the cheapest one too!
E = mc2
???
Angelo Mosso
Italian physiologist
( )
“[In Mosso’s experiments] the subject to be observed lay on a delicately balanced table which could tip downward either at the head or at the foot if the weight of either end were increased. The moment emotional or intellectual activity began in the subject, down went the balance at the head-end, in consequence of the redistribution of blood in his system.”
-- William James, Principles of Psychology (1890)

3. M R Timeline of MR Imaging 1920 1930 1940 1950 1960 1970 1980 1990
1972 – Damadian patents idea for large NMR scanner to detect malignant tissue.
Pauli suggests that nuclear particles may have angular momentum (spin).
1937 – Rabi measures magnetic moment of nucleus. Coins “magnetic resonance”.
1985 – Insurance reimbursements for MRI exams begin.
1973 – Lauterbur publishes method for generating images using NMR gradients.
MRI scanners become clinically prevalent.
1944 – Rabi wins Nobel prize in Physics.
1952 – Purcell and Bloch share Nobel prize in Physics. M R
NMR becomes MRI
1920
1930
1940
1950
1960
1970
1980
1990
2000
1973 – Mansfield independently publishes gradient approach to MR.
1990 – Ogawa and colleagues create functional images using endogenous, blood-oxygenation contrast.
1946 – Purcell shows that matter absorbs energy at a resonant frequency.
1959 – Singer measures blood flow using NMR (in mice).
1975 – Ernst develops 2D-Fourier transform for MR.
1946 – Bloch demonstrates that nuclear precession can be measured in detector coils.
FMRI – Week 1 – Introduction
Scott Huettel, Duke University

4. M R I f Timeline of MR Imaging 1920 1930 1940 1950 1960 1970 1980 1990
1972 – Damadian patents idea for large NMR scanner to detect malignant tissue.
Pauli suggests that nuclear particles may have angular momentum (spin).
1937 – Rabi measures magnetic moment of nucleus. Coins “magnetic resonance”.
1985 – Insurance reimbursements for MRI exams begin.
1973 – Lauterbur publishes method for generating images using NMR gradients.
MRI scanners become clinically prevalent.
1944 – Rabi wins Nobel prize in Physics.
1952 – Purcell and Bloch share Nobel prize in Physics. M R I
NMR becomes MRI f
1920
1930
1940
1950
1960
1970
1980
1990
2000
1973 – Mansfield independently publishes gradient approach to MR.
1990 – Ogawa and colleagues create functional images using endogenous, blood-oxygenation contrast.
1946 – Purcell shows that matter absorbs energy at a resonant frequency.
1959 – Singer measures blood flow using NMR (in mice).
1975 – Ernst develops 2D-Fourier transform for MR.
1946 – Bloch demonstrates that nuclear precession can be measured in detector coils.
FMRI – Week 1 – Introduction Scott Huettel, Duke University

5. The Rise of fMRI… Schleim & Roiser, 2009, Front. Hum. Neurosci.
Friston, 2010,
Science

6. … and the Decline of PET

7. FMRI – Week 1 – Introduction Scott Huettel, Duke University
Spatial vs. Temporal Resolution of Selected Brain Imaging Methods
FMRI – Week 1 – Introduction Scott Huettel, Duke University

8. The Brain Before fMRI (1957)
Polyak, in Savoy, 2001, Acta Psychologica
fMRI for Dummies

9. The Brain After fMRI (Incomplete)
reaching and pointing
motor
control
touch
eye
movements
retinotopic visual maps
grasping
executive control
motion
near head
memory
orientation selectivity
motion perception
moving bodies
social cognition
static
bodies
faces
objects
scenes 9

10. Magnetic Resonance Imaging Scanner

11. fMRI Setup
[Source: Mouser.com]

12. K-Space
Source: Traveler’s Guide to K-space (C.A. Mistretta)

13. K-Space Data gathered in k-space (Fourier domain of image)
transform ky kx
k-space
image space
Data gathered in k-space (Fourier domain of image)
Image is Fourier transform of acquired data
How k-space is sampled has implications for image

14. Partial k-space EPI Reduces TE (sacrifices some functional contrast)
EPI acquires one image
per TR
Due to symmetry, can
actually collect less than
full k-space ky kx
Reduces TE (sacrifices some functional contrast)
Must acquire slightly more than half (Hermetian symmetry is approximate)
Slight blurring added to image

15. Spiral FMRI Currently, only serious alternative to EPI
Short apparent TE (center of k-space acquired early)
Fast and efficient use of gradient hardware
Different artifacts than EPI (not necessarily better)

16. Parallel imaging (SENSE, SMASH, etc.)
Single coil
8-channel array
Surface coils
Coil sensitivity encodes spatial information
Can “leave out” large parts of k-space
Theory: For n coils, only need 1/n of k-space
Practice: Need at least ~1/3 of k-space
In general, incurs loss of SNR
More coverage, higher resolution, faster imaging, etc.

17. MR Safety Pacemaker malfunctions leading to death
At least 5 as of 1998 (Schenck, JMRI, 2001)
E.g., in 2000 an elderly man died in Australia after being twice asked if he had a pacemaker
Blinding due to movements of metal in the eye
At least two incidents (1985, 1990)
Dislodgment of aneurysm clip (1992)
Projectile injuries (most common incident type)
Injuries (e.g., cranial fractures) from oxygen canister (1991, 2001)
Scissors hit patient in head, causing wounds (1993)
Gun pulled out of policeman’s hand, hitting wall and firing
Rochester, NY (2000)
FMRI – Week 1 – Introduction
Scott Huettel, Duke University

18. MRI vs. fMRI MRI fMRI … low spatial resolution (~3-5 mm) high spatial
I is for Imaging
Imaging parameters are different MRI ”Anatomicals” aim to maximize tissue contrast
fMRI to maximize functional contrast. …
one 3D volume
(collected over several minutes)
series of 3D volumes (i.e., 4D data)
(e.g., every 2 sec for 5 mins)

19. PET and fMRI Activation
Source: Posner & Raichle, Images of Mind

20. fMRI Experiment Stages: Prep
1) Prepare subject
Consent form
Safety screening
Instructions and practice trials if appropriate
2) Shimming
putting body in magnetic field makes it non-uniform
adjust 3 orthogonal weak magnets to make magnetic field as homogenous as possible
3) Sagittals
Take images along the midline to use to plan slices
[Source: Wikipedia]

21. fMRI Experiment Stages: Anatomicals
4) Take anatomical (T1) images
high-resolution images (e.g., 0.75 x 0.75 x 3.0 mm)
3D data: 3 spatial dimensions, sampled at one point in time
64 anatomical slices takes ~4 minutes
64 axial slices
(3 mm)

22. Slice Terminology VOXEL (Volumetric Pixel) IN-PLANE SLICE
Slice Thickness
e.g., 6 mm
Gap, here 0 mm
Number of Slices
e.g., 10
Slice prescription (on SAG slice)
VOXEL
(Volumetric Pixel)
3 mm
6 mm
Matrix Size
e.g., 64 x 64
In-plane resolution
e.g., 192 mm / 64
= 3 mm
IN-PLANE SLICE
Field of View (FOV)
e.g., 19.2 cm

23. fMRI Experiment Stages: Functionals
5) Take functional (T2*) images
images are indirectly related to neural activity
usually low resolution images (e.g. here 3 x 3 x 6 mm)
all slices at one time = a volume (sometimes also called an image)
sample many volumes (time points) (e.g., 1 volume every 2 seconds for 136 volumes = 272 sec = 4:32)
4D data: 3 spatial, 1 temporal …

24. Region of interest (ROI)
fMRI Simplified
Time
fMRI
Signal
Intensity
ROI Time Course
Condition
~2s
Region of interest (ROI)
Condition 1
Time
Simplified fMRI experiment, we take multiple volumes of BASELINE, and then turn on some visual stimulation for the next 3 volumes CONDITION 2. repeat.
All fMRI data is: an intensity value of every voxel every time point , plot “Time Course”
perform our analysis on., and colorful blobs are what end up in publications
*When a brain region is working, it requires more oxygen, the signal in those tissues rises and falls in a way that is correlated to the testing paradigm….. As we can see in the time course
Condition 2
...
~ 5 min
Ignoring: HRF, subject motion, multiple comparisons

25. BOLD Time Course Blood Oxygenation Level-Dependent Signal
The metabolic signal we track is actually time-lagged from neural activity…
Positive BOLD response
Initial Dip
Overshoot
Post-stimulus Undershoot 1 2 3
BOLD Response (% signal change)
Time
Stimulus
“HRF”
(Hemodynamic
response
function)

26. The Canonical FMRI Experiment
Stimulus
pattern on
off
Predicted
BOLD signal
time
Subject is given sensory stimulation or task, interleaved with control or rest condition
Acquire timeseries of BOLD-sensitive images during stimulation
Analyse image timeseries to determine where signal changed in response to stimulation

27. BOLD/Signal Time Courses
Observed signal
(ARBITRARY UNITS)
MR SIGNAL
Predicted HRF
100*(y - baseline)/baseline
TIME
BOLD signal has arbitrary units:
varies from coil to coil, voxel to voxel, day to day, subject to subject

28. BOLD/Signal Time Courses
Observed signal
MR SIGNAL
(% Change)
Predicted HRF
100*(y - baseline)/baseline
TIME
Signal usually converted into units of % change:
Typically on the order of ½ - 4 %.

29. Final Statistics
For convenience/summarization, time course at each voxel can be converted to a scalar measure
Most common: parametric test of significance (e.g., t-test)
“statistical parametric map”: voxel-wise parametric test results

30. Stats on Anatomical

31. 2D  3D

32. Multiple Comparisons Problem
Voxel-level statistics assume independence of all voxels (not true)
Also, by virtue of the number of tests involved, conventional p-values are far too loose
p < 0.05 implies 5% chance of a false positive
Acceptable for one test, but with 100,000 tests (~ ½ brain size) that would be 5,000 false positives!…
Many options: Bonferroni (conservative), familywise error rate (FWE), false discovery rate (FDR), cluster level significance, and others

33. Limitations of Neuroimaging
Physical Limitations
spatial limitations (~1 mm)
temporal limitations (~50 ms to several seconds)
Physiological Limitations
noise
head motion
artifacts (respiration, cardiac pulse)
localization of BOLD response
vasculature
Current Conceptual Limitations
how can we analyze highly complex data sets?
brain networks
how are neural changes manifested in fMRI activation?

34. Limitations of Neuroimaging
Canonical cerebral microcircuit
(excitatory in red, inhibitory in black)
BOLD signal may
Increase or decrease,
and this doesn’t
necessarily tell what
the “neural input”
or “neural output”
was.
More complicated
than just “task
related neurons
firing”
Logothetis, 2008, Nature

35. BOLD Signal Dropout
BOLD
Non-BOLD
Dephasing near air-tissue boundaries (e.g., sinuses)
BOLD contrast (using long TE) coupled to signal loss (“black holes”)

36. Additional Topics Practical steps for image analysis (next)
Biological sources of the BOLD signal
Physics and underpinnings of MR signal
Computational background on image processing steps
Experimental design