Analysis of Temporal Relationships between ASD and Brain Activity through EEG and Machine Learning

Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder that impairs normative social cognitive and communicative function. Early diagnosis is crucial for the timely and efficacious treatment of ASD. The Autism Diagnostic Observation Schedule Second Edition (ADOS-2) is the current gold standard for diagnosing ASD. In this paper, we analyse the short-term and long-term relationships between ASD and brain activity using Electroencephalography (EEG) readings taken during the administration of ADOS-2. These readings were collected from 8 children diagnosed with ASD, and 9 low risk controls. We derive power spectrums for each electrode through frequency band decomposition and through wavelet transforms relative to a baseline, and generate two sets of training data that captures long-term and short-term trends respectively. We utilize machine learning models to predict the ASD diagnosis and the ADOS-2 scores, which provide an estimate for the presence of such trends. When evaluating short-term dependencies, we obtain a maximum of 56% accuracy of classification through linear models. Non-linear models provide a classification above 92% accuracy, and predicted ADOS-2 scores within an RMSE of 4. We use a CNN model to evaluate the long-term trends, and obtain a classification accuracy above 90%. Our findings have implications for using EEG as a non-invasive bio-marker for ASD with minimal feature manipulation and computational overhead.