Self-supervised transformer for time-series fMRI in autism detection

This is the code for Self-supervised transformer for time-series fMRI in autism detection.

Directory

• Pretrain: the directory contains the code for pretraining of the code using different random masking strategies.
• Code: the directory contains the code for downstream ASD classification task

Pretraining

• Three random masking tasks classes are in the python script: pretrain/utils/transform.py, and each is named as RandomMask, RandomMaskTime, RandomMaskROI
• To change different masking task, change line 132 in pretrain/utils/data_util.py to the class you want. The current implementation is RandomMaskROI
• To run the pretraining process, change the parser arguments in pretrain/runExp.sh, or pretrain/utils/parser_util.py, then run the following command:

bash pretrain/runExp.sh 

ASD Classification

• Change the parser arguments in code/runExp.sh or code/utils/parser_util.py to determine the file path, save path, and whether to utilize pretrained model or not
• ASD classification can be performed on ACE or ABIDE by running the following command for model training:

bash code/runExp.sh
bash code/runExpAce.sh

• the testing script is as follow:

bash code/runExp.sh
bash code/runExpAce.sh