Evaluating the Severity of Autism Spectrum Disorder from EEG: A Multidisciplinary Approach Using Statistical and...

A developmental impairment known as autism spectrum disorder (ASD) impacts youngsters and is characterized by impaired social communication and limited behavioral expression. In this study, electroencephalography (EEG) is used to obtain the brain electrical activity of typically developing children and of mild, moderate, and severe ASD patients using relative powers. This study investigates ASD patients using a multidisciplinary approach involving two-way ANOVA and Pearson's correlation statistical analyses to better understand the multistage severity of ASD from EEG by providing a spectro-spatial profile of ASD severity. Artificial intelligence frameworks, including a decision tree (DT) machine learning classifier and a long short-term memory (LSTM) neural network, are applied to discriminate mild, moderate, and severe ASD patients from typically developing children. The statistical results revealed that with increasing severity compared to the control, faster frequencies decreased and slower frequencies increased, indicating a distinct correlation between the severity of ASD and neurophysiological activity. Moreover, the DT classifier achieved a classification accuracy of 65%, and the LSTM classifier achieved a classification accuracy of 73.3%. This approach highlights the potential for statistical and artificial intelligence techniques to reliably identify EEG abnormalities associated with ASD, which could lead to earlier treatment and improved prospects for patients.