Autism Spectrum Disorder classification using EEG and 1D-CNN

Autism Spectrum Disorder (ASD), a neurodevelopmental disorder, is clinically diagnosed by behavioral assessment of the subjects. In this study, we used Electroencephalography (EEG) signals and Convolutional Neural Network (CNN) for ASD detection. We trained a multi-layered CNN using the EEG signals from both the autism and normal subjects. A test set which consisted of data from both autism and normal control subjects was used for evaluation of the model. We trained the model using the 5-fold validation technique. Batch normalization and dropout layers were used to avoid over-fitting. An accuracy of 0.922 was achieved using the proposed CNN. The convolution layers extract the features and the max pooling layers performed feature reduction. The increase in classification accuracy with the increase in number of subjects in training data was also observed. The results obtained in this study indicate that the Autism Spectrum Disorder can be detected using the EEG signals by automatic feature extraction and classification using the CNN.