lyman-tools

Tools for fmri analysis in the Lyman pipeline

This toolbox is intended to provide some helpful scripts for basic fMRI analyses, building off the Lyman pipeline. Some code is adapted from Lyman, and Ian Ballard.

ROI Tools

Create a spherical ROI from peak activation

create_sphere_frompeak.py: Take the peak MNI coordinate from group zstat1_localmax.csv, transform into avg152 coordinate space, create a sphere of sphere_radmm around the peak, and then mask with thresholded map. Output both the sphere and masked sphere niftis. This is useful to run before extracting parameter estimates from an ROI of interest to visualize the pattern of effect.

Extract parameter estimates from ROI

extract_copes.py: Given a csv file of masks, pull out the parameter estimates from subject-level fixed effects in either native or MNI space. Masks can be anatomical masks (e.g., generated by mask_masks.py), or defined by a contrast. This can also be used in conjunction with create_sphere_frompeak.py to pull parameter estimates from a spherical ROI defined from group results in MNI space.

Timeseries analyses

These analyses assume that your functional data has been processed with some variant of run_fmri.py -w preproc reg -t -u -reg epi, such that the unsmoothed raw data is preprocessed, and then coregistered into epi space (1st run) for each subject. An onset file, containing info about condition labels and onsets, is also necessary.

FIR model

run_fir.py: Extract timeseries for each subject, and use Nitime's EventRelatedAnalyzer to calculate the FIR event-related estimated of the HRFs for events of interest.

Extract timeseries, locked to event onset

run__extractraw.py: Extract "raw" (but preprocessed/realigned) timeseries, relative to event onsets. Integration (using the composite trapezoidal rule) is also performed over a specified window. This analysis is done on a trial-by-trial basis, so can be helpful for trial-wise correlations between regions or multivariate measures.

PPI analyses

(Adapted from code from Ian Ballard: https://github.com/iancballard/fd_fmri/blob/master/ppi_analysis.ipynb)

Generate design files for PPI analysis

run_create_ppidesign.py: Extract the coregistered preprocessed timeseries from a given ROI, take the principal eigenvariate, and z-score within run. Then convolve the psychological regressor with the HRF, center, and multiply with the timeseries to produce the interaction. Save out the timeseries and interaction as regressors, to be input as a regressor_file in the Lyman model parameters (in the experiment file).