EEG-Based Early Detection of Autism Using Convolutional Neural Networks

Abstract

This Timely intervention and improved management of autism spectrum disorder (ASD) are contingent upon early detection. In this paper a novel method for early autism identification using EEG signals and convolutional neural networks (CNNs) is proposed. Preprocessing, wavelet transform, Discrete Cosine Transform (DCT) , energy and entropy function feature extraction, and CNN classifier classification are some of the phases in the suggested approach. EEG signals are first pre-processed to get rid of artifacts and noise. Then, to extract pertinent characteristics from the EEG signals, wavelet transform and DCT are used. For feature extraction, energy and entropy calculations are used to identify unique patterns suggestive of ASD. After then, a CNN classifier receives these features and divides them into two categories: Autism identified or normal identified. The accuracy, specificity, sensitivity, and precision of the suggested approach are among the performance measures that are used to assess its effectiveness. With an accuracy of 92.34%, specificity of 92.95%, sensitivity of 92.65%, and precision of 92.65%, the experimental findings show encouraging performance. When compared to current systems, the suggested approach performs significantly better, outperforming the 91.78% accuracy of the current system.