rsfmri-classifiers

Autism classifiers with fMRI data

• This repo is intended as a demo on training / cross-validating models in Keras
• Uses publicly available MR scans from young children (2-5 years)
• Target is Autism diagnosis (binary class label)
• Features are timeseries of 200 regions
• Small sample size (<1000 scans), data quality is suspect...
• Explores linear, MLP, LSTM, and CNN models

To get the ABIDE-I data

• The data is public but requires registration
• Use the download script from this repo to get the fmri data. Note this repo is written in python2.7.
• And retrieve the phenotypic csv file here (you need to register)
• For more info about the ABIDE dataset, see here.
• Set the paths to the downloaded files in load_abide_data.py and run to generate the filtered and processed data

Build your model

• Train and validate your model in the notebook