A study on EEG feature extraction and classification in autistic children based on singular spectrum analysis method

Abstract

Autism spectrum disorder (ASD) is a serious neurodevelopmental disorder with core symptoms which include communication and social disorders and stereotyped behavior (Chez et al., 2006). The current incidence of ASD is rising rapidly around the world. The cause of ASD is presently unclear, and its diagnosis is mainly based on rating scales and behavioral observation, which are subjective. Therefore, it is worthwhile to find objective indicators for accurate assessment. Electroencephalography (EEG) is a powerful tool for analyzing and evaluating brain development due to its millisecond time resolution (Ingber & Nunez, 2011). The frequency domain of resting-state EEG recordings is generally divided into four bands: delta band, theta band, alpha band, and beta band (Akin, 2002). Alpha band oscillation is one of the most significant dynamic features of the brain in resting and awake states (Fan, 2018; Klimesch, Sauseng, & Hanslmayr, 2007). Most studies define alpha band as a fixed narrowband (commonly 8-12 Hz) (Haegens, Cousijn, Wallis, Harrison, & Nobre, 2014). Berger discovered alpha oscillation in 1929, and it has been a research topic in neuroscience ever since (Berger, 1929; Mierau, Klimesch, & Lefebvre, 2017). Previous studies have shown that alpha rhythms are mainly distributed in the parietal and occipital regions and are considered to be typical EEG waveforms of basal and resting EEG activity. Some research

Intellectually able adults with autism spectrum disorder show typical resting-state EEG activity

“…This is inconsistent with some of the studies using machine learning on resting-state EEG in ASD which reported classification accuracies above 87% 15 , 16 , 24 – 31 . However, the evaluation of the model performances were problematic in some of the studies.…”

“…However, increased alpha power 13 and gamma power have also been reported in ASD 14 . Other spectral features, e.g., peak alpha frequency 15 , theta/beta ratio 16 and asymmetry 17 have also been associated with ASD. Besides spectral features, abnormal functional connectivity 18 , microstates 19 , and measurements of criticality 20 , 21 have also been found in ASD.…”

“…Different models and features have been employed, e.g., discriminant function analysis using spectral power have yielded 90% accuracy 16 and using coherence 87% balanced accuracy was obtained 24 . K-nearest neighbor using alpha power resulted in 96.4% accuracy, and support vector machines (SVM) have also been utilized with alpha power and peak alpha frequency and achieved 93% accuracy 15 . SVM have also been employed on recurrence quantification analysis features, which resulted in 92.9% accuracy 25 , and on 20 features consisting of some descriptive statistics (e.g.…”

Predicting the Symptom Severity in Autism Spectrum Disorder Based on EEG Metrics

“…Compared with the magnetic resonance imaging, electroencephalography (EEG) has its tremendous advantages for exploring the network deficits occurring in ASD patients, as well as providing a reliable biomarker for characterizing ASD [3], [28]. EEG provides a convenient and easy-operating way, which is well tolerated by children with ASD particularly, for measuring and understanding both resting brain activity and task-engaged states in ASD patients.…”

Feature Extraction Method of Electronic Information Based on Statistical Correlation

“…e increasingly high-end technological achievements make electronic information technology to be one of the most dynamic science and technology in the 21st century, which permeates all aspects of social development and cannot be replaced. Speci cally, its application features are as follows [4].…”