Deep Learning Methods for Autism Spectrum Disorder Detection Using Functional MRI

Autism spectrum disorder (ASD) is among the most prevalent neurodevelopmental disorders, and traditional diagnostic methods are often slow and subjective. In this study we implemented six different deep learning architectures for detecting and classifying subjects with ASD using resting state functional magnetic resonance images (rs-fMRI). We tested the performance of the implemented deep learning models on 871 subjects from the publicly available Autism Brain Imaging Data Exchange (ABIDE) dataset. Functional connectivity matrices derived from rs-fMRI were used for training the models. The highest accuracy of 70.40% and precision of 71.55% were achieved by a model with three hidden layers, while the highest recall of 77.53% and Fl-score of 67.16% were achieved by a model with four hidden layers.