PSYC04H3 Lecture 2: Analysis of fMRI Data
6 Feb 2019
School
Department
Course
Professor
PSYC04: Brain Imaging Lab Clara Rebello
PSYC04 Lecture 2: Analysis of fMRI data
• DICOM: Digital Imaging and Communications in Medicine
o Has to be converted into a file that can be used for statistical analysis
• NIfTI: Neuroimaging Informatics Technology Initiative
o NifTI: volumetric data format
o (*.nii,*.hdr/*.img)
o Each space is found at a (x,y,z) coordinate
• Brain imaging data structure(BIDS): A format for organizing and describing outputs f
neuroimaging experiments
• SPM12: (Statistical Parametric Mapping) Software package designed for the analysis of brain
imaging data sequences
o Free and open source software (GPL)
o Requirement
▪ Matlab version must be at least 7.4
o Supported platforms
▪ Linuz
▪ Windows
▪ Mac
o Standalone version available
• Statistical parametric mapping: Refers to the construction and assessment of spatially extended
statistical processes used to test hypotheses about functional imaging data
• Roughly 3 stages to analyses
o Pre-processing: Multiple steps to prepare data for stats
o Statistical analyses: Lots of t-tests
o Post-stats: Thresholding of statistical results, locating where significant activity is in the
brain
• Steps to pre-processing
o Image reconstruction
o Slice-timing correction
▪ ShortTR not mandatory
▪ Use temporal derivatives
o Motion correction/realignment
▪ Always do this
o Undistortion of data
▪ We can also use motion parameters in states as covariates
o Corregistration (Structural/fMRI)
o Registration (normalization)/segmentation
o Spatial filtering
▪ Improve SNR, compensate inaccuracies in inter-subject alignment
o Temporal filtering
▪ Remove slow drifts
o Global intensity normalisation
▪ Keeps overall signal mean constant
PSYC04: Brain Imaging Lab Clara Rebello
• Data processing overview
o Take mean of data series
o Correct motion, realign, and fix distortion
o Use the anatomical data taken from the anatomical MRI in order to get estimate spatial
norm
o Registration/normalization of functional data, then spatially normalize it
• Motion correction
o fMRI data involves continuous scans (i.e. multiple volumes) across time
o People move in the scanner even with padding around head
o Scans und up shifting in location
o For statistics, each voxel needs to be located in the same anatomical place across time
o Motion correction aligns all acquired volumes to a common reference (Ex. 3rd volume)
• Rigid-body transformations
o Assume that brain of the same subject doesn’t change shape or size in the scanner
o Head can move, but remains the same shape and size
o Some exceptions
▪ Image distortions
▪ Brain slops about slightly because of gravity
▪ Brain growth or atrophy over time
o If the subject’s head moves, we need to correct the images
▪ Do this by image registration
• A 3-D rigid-body transformation is defined by
o 3 translations in X, Y, and Z directions
o 3 rotations about X, Y, and Z axes
• In 3-D rigid-body transformations
o Pitch corresponds to movement on x-axis
o Roll corresponds to movement on y axis
o Yaw corresponds to movement on z axis
Document Summary
Image reconstruction: slice-timing correction, shorttr not mandatory, use temporal derivatives, motion correction/realignment, always do this, undistortion of data, we can also use motion parameters in states as covariates, corregistration (structural/fmri, registration (normalization)/segmentation, spatial filtering. Improve snr, compensate inaccuracies in inter-subject alignment: temporal filtering, remove slow drifts, global intensity normalisation, keeps overall signal mean constant. 3rd volume: rigid-body transformations, assume that brain of the same subject doesn"t change shape or size in the scanner, head can move, but remains the same shape and size, some exceptions. Image distortions: brain slops about slightly because of gravity, brain growth or atrophy over time. In 3-d rigid-body transformations: pitch corresponds to movement on x-axis, roll corresponds to movement on y axis, yaw corresponds to movement on z axis. Clara rebello: motion estimates example, without motion correction (left) vs. with motion correction (right, corregistration. Inter-modal registration: match images from same subject but different modalities, anatomical localisation of single subject activations.
