Basic Units of Inter-Individual Variation in fMRI Data
Poster
1. ● With the group level analysis we obtained a brain image and for the group
brain activity when the tasks of looking at a face, cat, bottle, and chair.
● We had a threshold for the minimum z-statistic displayed (for every voxel)
for the image at 2.3. Voxels under the threshold were not highlighted
● We created contrasts to compare brain activity between tasks. These
contrasts were face - cat, face - bottle, face - chair, cat - face, bottle - face,
chair - face.
● Two step-process – Define clusters by arbitrary threshold 2.3 – Retain
clusters larger than alpha-level threshold k-alpha!
Big Data and Brain Imaging
Kaifeng Chen, Kunle Oshiyoye, Khuzaima Hameed, Xiaoting Sun, Xiaohe Yu
● Data collection - 6 subjects collected from openfMRI using
four tasks (comparisons of brain images), with 12 runs for
each subject
○ each run consisted of subjects looking at images of the
following objects:
■ face
■ cat
■ bottle
■ chair
○ brain images were constructed from magnetic signal data
● Using box-car
● Used FSL to do first-level and higher-level analysis to
preprocess and analyze the raw data
Data
face - cat face - bottle
face - chair face - cat cat - face
combination of face - cat, face - bottle and face - chair
Is There Any Specific Brain Region Unique for Facial Recognition?
Kunle Oshiyoye1, Khuzaima Hameed2, Kaifeng Chen2, Xiaohe Yu2, Xiaoting Sun2
1Eastern Michigan University, 2University of Michigan
● A majority of PPA area was more activated when the subject was
exposed to a cat or a chair than to a face. However, such area was
not obvious in the face-bottle comparison.
● Part of FFA area was more activated when the subject was exposed
to a face than to a bottle and a chair. Nevertheless, such area was
not obvious in the face-cat task.
● Our results seem to support the second theory, and some
inconsistencies could be due to small sample size (six subjects)
[1] Maura L. Furey. Alumit Ishai. Jennifer L. Schouten. Pietro Pietrini. James V. Haxby and M. Ida Gobbini.
Distributed and overlapping representations of faces and objects in ventral temporal cortex. Science,
293(5539):2425-2430, 2001
ResultsIntroduction
● Motivations - infer regions of brain where facial recognition
differs from object recognition
○ FFA - fusiform face area: particular area in brain
postulated to perform facial recognition tasks
○ PPA - parahippocampal place area used for memory,
encoding and retrieval, also for scene recognition
○ three theories of brain activity for facial recognition
1. specific region/regions unique to facial recognition
● FFA
2. specific region/regions that also perform other
perceptive tasks
● FFA and PPA
3. wide overlapping/interconnected regions
● fMRI - functional magnetic resonance imaging, used for
researching brain activity
○ made up of voxels: volumetric pixels
○ voxel color represents intensity of magnetic signal
● BOLD - blood oxygenation level dependent
○ hemodynamic response
○ signal intensity increases with increased metobolic
demands of neurons
○ ratio of oxyhemoglobin to deoxyhemoglobin changes
BOLD response
Methods & Analysis
We use FSL to analyze the resulting data.
○ it shows us the activity (BOLD response) for every area in
the brain.
○ BOLD activity is represented with colors, the lighter the
color the greater the intensity.
● We combine the responses across all of the subjects to create a
group level (mixed effects) analysis
● FSL was used for preprocessing, techniques used included:
○ Slice-time correction: interpolate data to match the timing
of the chosen reference slice
○ Motion correction: adjust the image to reduce errors caused
by the head motion happened during data acquisition
○ Co-registration: register all subjects’ images to a standard
brain template
○ Spatial smoothing: blur the image to remove high-frequency
noise and to reduce mismatch in signal location across
individuals
○ Temporal filtering: remove low frequency drift in time
series
● Model fitting
○ GLM - Generalized Linear Model
● the data for a single subject at voxel j can be written as
yj = Xjβj + εj , with εj ∼ N(0, V)
Conclusions
References
● Contrasts of face with chair, bottle and cat were obtained from higher level analysis (group level
analysis)
● Contrasts of significance were determined by cluster level thresholding using random field theory
○ Two step-process
– First, select all voxels whose standardized contrast greater than a threshold
– Second, retain clusters of voxels whose volume is greater than a second threshold
● Data collection - 6 subjects collected from openfMRI
using four tasks (comparisons of brain images), with 12
runs for each subject
○ each run consisted of subjects looking at images of the
following objects:
■ face
■ cat
■ bottle
■ chair
○ Statistical parametric maps of brain activity were
constructed from fMRI data
● Using box-car design (block design)
● Used FSL (1http://fsl.fmrib.ox.ac.uk/) to do first-level and
higher-level analysis to preprocess and analyze the raw
data
face - cat face - bottle face - chair
combination of cat - face,
bottle - face and chair - face
combination of face - cat,
face - bottle and face - chair Thanks to Professor Timothy Johnson and GSI Zhuqing Liu for providing
guidance and care and University of Michigan School of Public Health for
the free boxed lunch
Acknowledgements
2.3 4.12.3 3.5 2.3 4.3
2.3 4.02.3 3.2
Data
2.3 4.3
cat - face bottle - face chair - face