Manuscript received April 1, 2026; revised May 9, 2026; accepted July 13, 2026; published November 14, 2026.

Abstract—Autism Spectrum Disorder (ASD) is a neurodevelopmental condition whose diagnosis still relies largely on behavioral assessment, creating a need for more objective computer-aided screening approaches. This study proposes a classification framework based on multi-band electroencephalography connectivity matrix images derived from resting-state recordings. Electroencephalography signals from 16 subjects, consisting of 8 subjects with autism spectrum disorder and 8 typically developing controls, were preprocessed using Butterworth bandpass filtering and independent component analysis. The denoised signals were then decomposed into theta, alpha, and beta bands, segmented using overlapping windows, and transformed into phase locking value connectivity matrices. These matrices were converted into connectivity images and classified using two deep learning architectures, namely Data-efficient Image Transformer and ResNet18, under subject-wise 8-fold stratified cross-validation. In the single-band setting, the beta-band images yielded the best performance for both models, achieving mean accuracies of 0.9817 ± 0.0032 for Data-efficient Image Transformer and 0.9689 ± 0.0079 for ResNet18. In the fusion setting, combining theta, alpha, and beta connectivity images further improved performance, producing the highest mean accuracies of 0.9897 ± 0.0048 for Data-efficient Image Transformer and 0.9860 ± 0.0041 for ResNet18. Wilcoxon signed-rank analysis confirmed that the fusion schemes significantly outperformed the best single-band baseline. These findings demonstrate that multi-band electroencephalography connectivity image fusion provides an effective representation for autism spectrum disorder classification and that transformer-based modeling offers a consistent advantage over the convolutional baseline.

Keywords—autism spectrum disorder, electroencephalography, functional connectivity, phase locking value, connectivity matrix image, multi-band fusion, deep learning

Cite: Zulfan Zulfan, Junidar Junidar, Kikye Martiwi Sukiakhy, Aufa Rafiki, and Maiza Duana, "Fusion of Multi-Band EEG Connectivity Matrix Images for Autism Spectrum Disorder Classification Using Transformer and CNN Models," Journal of Image and Graphics, Vol. 14, No. 6, pp. 997-1014, 2026.

Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).