Deep-Fusion of Scalogram and Spatio-Temporal EEG Features with Attention Mechanism for Autism Spectrum Disorder...

Autism Spectrum Disorder is a neurodevelopmental condition characterised by social and communication challenges, repetitive behaviours, and sensory sensitivities. Early diagnosis allows for timely intervention, improving quality of life. This study introduces a novel approach to Autism Spectrum Disorder detection using resting-state electroencephalography (EEG), which captures neuronal activity to identify Autism Spectrum Disorder-related abnormalities. Our methodology integrates spatiotemporal EEG features with image-based features extracted from scalograms. To improve classification performance, we proposed a novel deep-learning pipeline that combined scalogram and EEG signal features to emphasise feature fusion. This fusion was processed through an EfficientNet-based model, enhanced with a Convolutional Block Attention Mechanism to capture comprehensive spatial and temporal representations. The attention module leveraged both channel and spatial attention to refine feature extraction. Our model achieved an accuracy of 84%, with an F1-score of 82%, recall of 82%, precision of 82%, and an AUC-ROC of 0.89, demonstrating the potential of EEG in distinguishing ASD from healthy controls.Clinical relevance-EEG offers a non-invasive, cost-effective, and objective approach to early Autism Spectrum Disorder detection. Unlike subjective behavioural assessments, it provides direct neural insights, enabling scalable screening, especially in young children. This study enhances the diagnostic accuracy of Autism Spectrum Disorder by integrating single channel scalogram and spatiotemporal-based deep feature fusion with attention mechanism, establishing EEG signatures as a promising biomarker and a valuable clinical tool.