Detection of Autism Spectrum Disorder from EEG signals using pre-trained deep convolution neural networks

Electroencephalography (EEG) is a method of recording the electrical activity of the brain and it has been used to detect various neurological disorders which are associated with the abnormal brain electrical activity. Various diseases such as epilepsy and Alzheimer's disease are being detected using EEG. Autism spectrum disorder (ASD), which is neurological disorder impairs the socialization, communication and behavior of the subjects suffering from it. Various studies have been conducted to find the EEG abnormalities in ASD subjects and normal controls, so that ASD can be detected using EEG. Machine learning and deep learning networks have been used to classify normal and abnormal EEG waveforms and several studies have used Convolution neural networks (CNN) for the classification. In this study, we have used transfer learning approach to train the pre-trained CNNs, GoogLeNet and SqueezeNet for classifying ASD subjects and normal controls using their EEG signals. The accuracy achieved using the GoogLeNet and SqueezeNet were 75% and 82% respectively in classifying the scalograms generated from EEG signals of ASD subjects and normal control subjects. The results obtained indicate that this method can assist in classifying ASD subjects and normal subjects using EEG signals.