Skip to content

Latest commit

 

History

History
34 lines (19 loc) · 2.16 KB

README.md

File metadata and controls

34 lines (19 loc) · 2.16 KB
Raw

Decoder Construction for rt-fMRI Image Evolution

Under construction! This project is in the pilot phase, and may not work out-of-the-box on a new machine with new data.

This repo contains the python code for creating model files for brain decoding of fear ratings. Input data from a Siemens machine are converted to NIfTI, and then BIDS format. There is then a script for image alignment in native space, followed by a Jupyter notebook for computing the Principal Component Analysis and logistic regression. There are also the psychopy programs for video presentation and ratings, as well as the video stimuli.

convert_to_nifti.py

Takes a directory of .DCM or .IMA files and converts them to NIfTI formatted files using dcm2niix.

convert_to_bids.py

Reorganizes and renames the NIfTI files using BIDS community conventions.

alignment.py

Script for aligning fMRI frames to the anatomical image. Uses the ANTSpyx library, for fast affine registration. Manual alignment, rather than a preprocessing library like fMRIPrep, is needed in order to ensure that offline and online preprocessing steps are identical. Preprocessing also needs to be minimal to ensure that single frames can be processed in less than the time-length of one frame acquisition (TR).

model.ipynb

Jupyter notebook that collects the fMRI and events data, and computes the PCA and logistic regression models used for realtime decoding.

Outputs are:

  • Brain mask in native space (based on anatomical T1w scan)
  • NiftiMasker object (nilearn) for converting fMRI runs to 2D matrices
  • PCA model
  • Logistic regression models for fear, disgust, and cuteness predictions

Psychopy

In the psychopy directory are two programs, one for online video presentation and one for offline (out-of-scanner) video rating. The video rating program is also adapted for web-based completion via Pavlovia, using the .js version of the program.

Two stimulus sets, involving 200 animal videos comprised of 10 unique categories each, are also included. These videos are acquired with a Creative Commons license from websites including Pexels, Pixabay, Mixkit, and YouTube. Videos were processed by cropping to 3s in length and 768*512 pixels.