Iterative consensus spectral clustering for brain functional modules

The code repo for the iterative consensus spectral clustering that performs detection of subject and group level brain functional modules. We obtained the following group level modularizations for normal healthy samples:

The proposed algorithm can also be used to obtain subject-level modules from multiple scans (of the same subject). For more details refer to the paper.

Setup

pip3 install -r requirements.txt

Guide to run the code

1. The data for the subjects resides in the data directory. You can create subdir for the dataset in the data directory and put the connectivity matrices for each subject. In the paper, we used the HCP dataset to obtain our results. For your reference, we have added some dummy connectivity matrices (synthetic connectivity matrices with 264 nodes) to give an idea of the format we expect the subject data to be in. The data should be a numpy array for each subject, of the shape (N, N) where N is the number of regions of interest. Currently, we have:

• data/multiple_subjects/ folder containing data for 25 synthetic samples in the form of subject_*.npy. Each sample is a numpy array of shae (264, 264) that denotes the adjacancy matrix for the functional connectome of the subject. You can use this data to obtain group-level modularizations (and individual-level modularizations).

• data/subject_sessions/ folder containing data for 2 subjects, containing 64 synthetic scan session samples in the form of data/subject_sessions/subject_*/S*_corr.npy. Each sample is a numpy array of shae (264, 264) that denotes the adjacancy matrix for the functional connectome of the subject's scan. You can use this data to obtain subject-level modularizations (and session/individual-level modularizations).

2. The codes reside in the codes folder. We give two scripts to derive group-level modularizations and subject-level modularizations.

• codes/group_level/ folder contains the code to obtain group-level modularizations from subject connectivity matrices. As a starting point we have added dummy data in data/multiple_subjects/. The code to derive group-level modularizations (and other metrics) can be run by

  python3 ICSC_group_level.py
  

• codes/subject_level/ folder contains the code to obtain subject-level modularizations from multiples scans of the same subject. As a starting point we have added dummy data in data/subject_sessions/subject_*/S*_corr.npy. The code to derive subject-level modularizations (and other metrics) can be run by:

  python3 ICSC_subject_level.py
  

You can modify the following parameters in ICSC_subject_level.py or ICSC_group_level.py files to modify the runs as per your requirement:

• NUM_NODES: the number of ROIs considered
• NUM_RUNS: the number of independent runs you wish to consider
• NUM_THREADS: the number of cores you want to use. Is useless if NUM_RUNS < 2
• DATASET: the folder where the dataset is stored.
• MAX_LABELS, MIN_LABELS: The range for the number of modules for the subject/scans under considerations
• SAVE_DIR: The directory where the results will be saved.

If you use this algorithm/code please cite our paper: Sukrit Gupta, and Jagath C. Rajapakse. 'Iterative consensus spectral clustering improves detection of subject and group level brain functional modules', Nature: Scientific Reports