Diagnosing Autism Spectrum Disorder in Pediatric Patients via Gait Analysis using ANN and SVM with Electromyography Signals

Abstract

Autism Spectrum Disorder (ASD) is a permanent neurological maturation condition that impacts communication, social interaction, and behavior. It is also associated with atypical walking patterns. This study aims to create an automated classification model to distinguish ASD children during walking based on the muscles Electromyography (EMG) signals. The study involved 35 children diagnosed with ASD and an equal number of typically developing (TD) children, all aged between 6 and 13 years. The Trigno Wireless EMG System was used to collect EMG signals from specific muscles in the lower limb (Biceps Femoris -BF, Rectus Femoris -RF, Tibialis Anterior -TA, Gastrocnemius -GAS) and the arm (Biceps Brachii -BB, Triceps Brachii -TB) on the left side. To identify the most significant features influencing walking in ASD children, a statistical analysis using the Mann-Whitney Test was conducted. The dataset contained 42 features derived from the analysis of six muscles across seven distinct walking phases throughout a single gait cycle. Following this, the Mann-Whitney Test was utilized for feature selection, uncovering five significantly distinctive features within the EMG signals between children with ASD and those who were typically developing. The most notable EMG features were subsequently employed in constructing classification models, namely an Artificial Neural Network (ANN) and a Support Vector Machine (SVM), aimed at distinguishing between children with ASD and those who were typically developing. The results indicated that the SVM classifier outperformed the ANN classifier, achieving an accuracy rate of 75%. This discovery shows potential for employing EMG signal analysis and classification model algorithms in diagnosing autism, thereby advancing precision health.