Neural EEG Signatures Of ASD In Children. LRNN Or NARXNN Along With Linear Prediction Coefficient Or Levinson Durbin Recursion Feature Extraction Modelsfor The Early Diagnosis Of ASD

Abstract

Autism spectrum disorder (ASD) as the name suggests has a wide variety of spectrum of clinical features and is very difficult to diagnose early in children. It is not viable to appoint neurologists to screen for this disease since 2.5crores children are born every year in India. There is a need to create cheaper and effective diagnostic techniques to diagnose this disease early. This study strives to close that gap using EEG (Electroencephalogram) which is a cheaper method than others to diagnose autism. Usable features from EEG were extracted using techniques like Linear prediction coefficient and Levinson Durbin Recursion feature extraction models and these features were classified using neural networks. This study showed a higher diagnostic accuracy of 93% and 96% respectively for LRNN(Layered Recurrent Neural Network) and NARXNN(Non-Linear autoregressive with external input neural network) methods..