Identification of Children With Autism Spectrum Disorder Based on Multidimensional EEG Feature Fusion Across Temporal-Spectral-Spatial Domains

Le résumé fourni par la source

Background: To better characterize the complex neural features of autism spectrum disorder (ASD) and overcome the limitations of traditional electroencephalography (EEG) analysis methods, we developed a multi-metric EEG framework integrating temporal, spectral, and spatial dimensions, systematically characterized the dynamics, individualization, and nonlinear network features of neural oscillations in ASD, and evaluated their classification performance. Methods: = 44) were recruited and resting-state EEG data were collected. The analysis was conducted from three perspectives: (1) temporal domain - Lempel-Ziv complexity (LZC) was used to quantify the dynamic complexity of signals; (2) frequency domain - the gedBounds method based on generalized eigen decomposition (GED) was applied to identify individualized frequency bands; (3) spatial domain - Generalized Symbolic Nonlinear Granger Causality (GSNGC) was used to construct brain functional networks and compute graph-theoretic metrics. Finally, a support vector machine (SVM) integrated multidimensional features for ASD classification. Results: In the temporal domain, the ASD group showed significantly lower whole-brain LZC compared with the TD group, with the most pronounced reduction observed in the alpha band, suggesting reduced neural dynamic information processing capacity. In the frequency domain, the ASD group showed an expanded theta bandwidth, reduced low-frequency power in central-occipital regions, and increased beta power in frontal regions. In the spatial domain, children with ASD exhibited an atypical connectivity pattern characterized by increased low-frequency connectivity, reduced alpha-band connectivity, and increased beta-band connectivity, along with significantly higher global efficiency in theta and beta networks. The SVM model integrating temporal, frequency, and spatial features achieved an accuracy of 89.2%, significantly outperforming single-domain feature models, confirming that multidimensional feature integration improves classification performance. Conclusions: This study introduces a novel analytical approach combining individualized frequency band identification, nonlinear connectivity modeling, and dynamic complexity analysis. The findings comprehensively reveal multi-scale abnormalities of neural oscillations in children with ASD and demonstrate the discriminative power of multi-dimensional EEG feature integration for ASD classification and auxiliary diagnosis, thereby providing a scientific basis for clinical diagnosis and intervention. Clinical Trial Registration: No: ChiCTR2400092790. 24 November, 2024, https://www.chictr.org.cn/showproj.html?proj=249950.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.