Autism spectrum disorder detection technique using EEG and convolution neural networks

• Published in: AIP conference proceedings Vol. 2603; no. 1
• Main Authors: Mohi-ud-Din, Qaysar, Jayanthy, A. K.
• Format: Journal Article Conference Proceeding
• Language: English
• Published: Melville American Institute of Physics 25.04.2023
• Subjects: Artificial neural networks; Autism; Brain; Continuous wavelet transform; Electroencephalography; Epilepsy; Medical imaging; Neural networks; Wavelet transforms
• ISSN: 0094-243X; 1551-7616
• DOI: 10.1063/5.0128348

Autism Spectrum Disorder (ASD) approximately affects 1 in 54 children in USA. Unfortunately, there is no cure for ASD and ASD subjects may require life-long assistance. Studies show that ASD is caused by abnormal brain development and the abnormal brain development gives rise to behaviors that are present in ASD subjects, which become the basis for clinical diagnosis of ASD. There is a time gap between abnormal brain development and the ASD behavioral development, which can cause delay in early intervention of ASD. Electroencephalograph (EEG) signals are utilized to detect neurological signals including epilepsy. In this study, we propose to use Convolution Neural Network (CNN) for ASD detection using the scalogram images created from EEG signals. The EEG data of Autism subjects and typical control subjects was segmented. We then used Continuous Wavelet Transform to generate scalograms from the EEG segments. We trained the model using scalogram images and obtained an accuracy of 77%. This is a first such approach of detecting ASD using scalogram images and CNN. We used a simple CNN network with fewer parameters. The results stipulate this technique can assist clinicians in ASD detection and the useof EEG for ASD detection can make early intervention possible, as we don’t have to wait for the behavioral assessment of the subjects.