A study on EEG feature extraction and classification in autistic children based on singular spectrum analysis method
Jie Zhao, Jiajia Song, Xiaoli Li, Jiannan Kang
Abstract
INTRODUCTION: The clinical diagnosis of Autism spectrum disorder (ASD) depends on rating scale evaluation, which introduces subjectivity. Thus, objective indicators of ASD are of great interest to clinicians. In this study, we sought biomarkers from resting-state electroencephalography (EEG) data that could be used to accurately distinguish children with ASD and typically developing (TD) children. METHODS: We recorded resting-state EEG from 46 children with ASD and 63 age-matched TD children aged 3 to 5 years. We applied singular spectrum analysis (SSA) to the EEG sequences to eliminate noise components and accurately extract the alpha rhythm. RESULTS: When we used individualized alpha peak frequency (iAPF) and individualized alpha absolute power (iABP) as features for a linear support vector machine, ASD versus TD classification accuracy was 92.7%. CONCLUSION: This study suggested that our methods have potential to assist in clinical diagnosis.