Skip to content

The code used for my Master's thesis: "Detecting mild traumatic brain injury from magnetoencephalography data using machine learning"

Notifications You must be signed in to change notification settings

veeraita/thesis-detecting-mtbi

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

53 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Detecting mild traumatic brain injury from MEG data using machine learning

This repository contains the code used for the Master's thesis by Veera Itälinna.

Repository structure

  • dicom2nifti

    Contains a MATLAB script for converting DICOM files to NIfTI format.

  • pipeline-tbi

    Contains the steps of the source modelling pipeline (adapted from Rantala, 2020).

  • model

    Contains the code for training and evaluating the model and visualizing the results.

The source modelling pipeline

  1. Run FreeSurfer on the MRI images (pipeline-tbi/freesurfer)
  2. Make BEM meshes with Watershed algorithm (pipeline-tbi/watershed)
  3. Perform ICA on the raw MEG measurements, and apply the ICA solutions also to the empty room recordings (pipeline-tbi/ica)
  4. Calculate noise covariance matrices from the empty room recordings (pipeline-tbi/noisecov)
  5. (Optional: Increase the dataset size using a sliding window approach on the MEG measurements (pipeline-tbi/window))
  6. Compute the coordinate transformations (pipeline-tbi/trans)
  7. Compute source PSDs and morph to an average brain (pipeline-tbi/psd)

Creating the dataset used in the thesis

  1. Calculate mean and standard deviation matrices from the normative samples (pipeline-tbi/averages)
  2. Calculate Z-score maps (pipeline-tbi/zmap)
  3. Parcellate the Z-maps (pipeline-tbi/parc)
  4. Collect all parcellated Z-maps into a single dataset (pipeline-tbi/zmap/create_zmap_dataset.py)

Training and evaluating the model

Training and evaluating the model is done by running the Python script model/run_zmap_svm.py with appropriate options.

usage: run_zmap_svm.py [-h] [-f FIT_PARAMS] [-g] [-r] [-n] [--fs]
                       [--norm-data {full,age,random}] [-v] [-p]

optional arguments:
  -h, --help            show this help message and exit
  -f FIT_PARAMS, --fit-params FIT_PARAMS
                        Parameters to pass to the classifier (JSON string)
  -g, --grid            Use grid search for model selection
  -r, --repeat          Use repeated cross-validation
  -n, --nested          Use nested cross-validation for model selection and
                        validation
  --fs                  Apply feature selection
  --norm-data {full,age,random}
                        Select what normative data to use
  -v, --visualize       Visualize the results
  -p, --perm-test       Use permutation test

How to obtain the results presented in the thesis:

  • Nested and repeated cross-validation with full normative data and all features: run_zmap_svm.py -n -r
  • Nested and repeated cross-validation with full normative data and feature selection: run_zmap_svm.py -n -r --fs
  • Nested and repeated cross-validation with age-matched normative data and all features: run_zmap_svm.py -n -r --norm-data age
  • Nested and repeated cross-validation with age-matched normative data and feature selection: run_zmap_svm.py -n -r --fs --norm-data age
  • Nested and repeated cross-validation with random normative data and all features: run_zmap_svm.py -n -r --norm-data random*
  • Nested and repeated cross-validation with random normative data and feature selection: run_zmap_svm.py -n -r --fs --norm-data random*

* repeated three times with different random seeds and averaged the results

References

Rantala, A. (2020). Creating a normative database of resting-state brain activity from a large number of MEG recordings [Aalto University School of Science]. http://urn.fi/URN:NBN:fi:aalto-202101311732

About

The code used for my Master's thesis: "Detecting mild traumatic brain injury from magnetoencephalography data using machine learning"

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published