Applying TCN-Based Rhythmic Electroencephalogram Data to Assist in the Construction of Autism Intervention Pathways in Music Therapy

This study addresses the challenge of personalizing music therapy pathways for autism spectrum disorder (ASD) by developing a rhythmic electroencephalogram (EEG) decoding framework based on Temporal Convolutional Networks (TCN). Considering EEG as a representative bioelectromagnetic signal, the proposed approach provides an effective means for extracting neural dynamics relevant to individualized intervention assessment. Pre- and post-intervention EEG data were collected from 55 children with ASD during a 12-week music therapy program. The TCN model achieved an accuracy of 0.861 (F1 = 0.843) in classifying treatment responders, outperforming baseline models including Transformer (accuracy = 0.816). For regression, it attained an R2 of 0.692 (RMSE = 6.52) in predicting changes in the Social Responsiveness Scale (∆SRS). Responders exhibited significantly higher Beta/Gamma relative power and Theta/ Alpha-weighted Phase Lag Index (wPLI). A Neural Response Index (NRI), derived from TCN-decoded features, showed a strong correlation with ∆SRS (r = 0.712, p < 0.001). Empirical application of NRI-based decision rules yielded an 88.9% response rate in the high-NRI pathway. These findings demonstrate that TCN can effectively decode bioelectromagnetic neural patterns associated with therapeutic outcomes and that NRI provides a quantitative, data-driven indicator for personalized music therapy planning, offering methodological insights for advanced electrophysiological signal analysis and intelligent biomedical sensing applications.