Deep neurocomputational fusion for ASD diagnosis using multi-domain EEG analysis

Autism spectrum disorder (ASD) presents significant challenges in early detection due to its heterogeneous nature and the subtlety of neurophysiological variations. Electroencephalography (EEG) has emerged as a promising tool for ASD diagnosis, offering insights into intricate neurophysiological sig...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Neurocomputing (Amsterdam) Jg. 641; S. 130353
Hauptverfasser: Rasool, Abdur, Aslam, Saba, Xu, Yongjie, Wang, Yishan, Pan, Yi, Chen, Weiyang
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Elsevier B.V 07.08.2025
Schlagworte:
Autism spectrum disorder
Deep learning
EEG analysis
Fusion model
Neurocomputational diagnostics
Deep learning
Fusion model
Neurocomputational diagnostics
Autism spectrum disorder
EEG analysis
ISSN:0925-2312
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Autism spectrum disorder (ASD) presents significant challenges in early detection due to its heterogeneous nature and the subtlety of neurophysiological variations. Electroencephalography (EEG) has emerged as a promising tool for ASD diagnosis, offering insights into intricate neurophysiological signal dynamics. However, existing computational EEG analysis techniques often struggle to effectively capture complex ASD-related neural activity patterns. To address these limitations, this study proposes an Encoder-Ensemble Fusion Model (EEFM), designed to enhance EEG-based ASD classification through advanced neurocomputational modeling. EEFM integrates multi-domain EEG features with a sophisticated fusion network, initially extracting a diverse range of time and frequency-domain characteristics. This model enables the identification of nuanced and heterogeneous ASD-specific neural activity, leveraging ensemble learning and autoencoder architectures to uncover complex relationships within the data. Specifically, an LSTM (Long Short-Term Memory)-autoencoder models temporal and spatial dependencies, followed by an XGBoost Regressor for nonlinear EEG signal interpretation and a linear regression model for decision boundary refinement. The proposed model has been evaluated on a dataset focusing on sex-specific EEG variations in ASD and Typically Developing (TD) subjects. By incorporating EEG-based neurophysiological biomarkers, EEFM achieved 93% accuracy in distinguishing ASD from TD individuals. This model enhances sensitivity to subtle ASD-related abnormalities while addressing challenges in data heterogeneity, scarcity, and generalizability, contributing to advancements in neurocomputing methodologies for EEG-based ASD diagnosis.
ISSN:0925-2312
DOI:10.1016/j.neucom.2025.130353