Detection of abnormalities based on gamma wave EEG signal for autism spectrum disorder

Abstract

Diagnosing Autism Spectrum Disorder (ASD) by using the traits of abnormalities in their gamma waveform has been proposed in this study to suggest an objective method to detect the disorder using Electroencephalography (EEG) signal. Gamma waveform plays an important role in learning, memory and information processing where it shows slower activities in ASD person compared to a normal person, thus, causing the patients to have trouble in processing knowledge, communicate and pay attention. This study applies Probabilistic Neural Network (PNN) and General Regression Neural Network (GRNN) to classify the data into normal and abnormal classes. Classification algorithm by PNN was used as a benchmark for the outcomes. The results show that even though PNN and GRNN have similar architecture, but with fundamental difference, the outcomes are different. In this case, PNN performs better than GRNN. To obtain the desired results, we used three and four statistical features (mean, minimum, maximum and standard deviation) for both methods. The outcomes of using PNN with four features are more accurate (99.5% for normal class and 80.5% for abnormal class) compared to only three features. Furthermore, the outcomes of using GRNN with four features also have improvement (95% for normal class and 63.5% for abnormal class) compared to only three features.