DiSCo Challenge repo for the SCIL team

This is the main repository of the processing pipeline for the SCIL's submission to the MICCAI CDMRI 2021 DiSCo Challenge.

Data

All of the relevant data is stored on braindata (braindata/datasets/DiSCo) and uses the BIDS format. Three datasets are available:

sub-DiSCo1 is the training data, for which we have the ground-truth connectivity matrix.
sub-DiSCo3 is the validation data, for which we also have the ground-truth connectivity matrix.
sub-DiSCo2 is the test data, for which we do not have the ground-truth connectivity matrix.

Pipeline

The pipeline does some very basic processing steps right now:

Step 1 : Denoise DWI
Step 2 : Extract shell used for WM and frf computing
Step 3 : Compute FA map
Step 4 : Compute tissue masks
Step 5 : Compute frf
Step 6 : Compute fodf
Step 7 : Compute local tracking (use 'local') OR Compute pft (use 'pft')
Step 8 : Run commit
Step 9 : Decompose connectivity
Step 10 : Compute connectivity
Step 11 : Compute correlation
Step 12 : Compute binary correlation
Step 13 : Compute confusion matrix

Getting started

You will need to have scilpy scripts loaded as part of your environment. To run the base version of the pipeline on the training data, you can run:

$ cd code
$ bash -x DiSCo_connectivity_pipeline.sh ~/braindata/databases/DiSCo/sub-DiSCo1/sub-DiSCo1_DWI_RicianNoise-snr30.nii.gz ~/braindata/databases/DiSCo/sub-DiSCo1/sub-DiSCo1_DWI_RicianNoise-snr30.bval ~/braindata/databases/DiSCo/sub-DiSCo1/sub-DiSCo1_DWI_RicianNoise-snr30.bvec ~/braindata/databases/DiSCo/sub-DiSCo1/sub-DiSCo1_ROIs.nii.gz ~/braindata/databases/DiSCo/sub-DiSCo1/sub-DiSCo1_Connectivity_Matrix_Cross-Sectional_Area.txt pft training

which should net you an r coefficient of around ~0.80.

Participating

To participate, simply fork this repo and start hacking ! You can add, remove or modify steps of the main pipeline as you wish.

Other teams' members