A Novel EEG-Based Topographic Brain Map–Driven Deep Learning Method for Autism Spectrum Disorder Detection in Children
Brain Topography, cilt.39, sa.5, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 39 Sayı: 5
- Basım Tarihi: 2026
- Doi Numarası: 10.1007/s10548-026-01243-1
- Dergi Adı: Brain Topography
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, BIOSIS, EMBASE, MEDLINE, Psycinfo, Natural Science Collection (ProQuest), Biological Science Database (ProQuest), Biomedical Reference Collection: Corporate Edition (EBSCO), Health Research Premium Collection (ProQuest), Pharma Collection (ProQuest)
- Anahtar Kelimeler: Autism spectrum disorder, Convolutional neural networks, Cross-dataset validation, Electroencephalography, Feature fusion, Topographic brain maps
- Çukurova Üniversitesi Adresli: Evet
Özet
Autism Spectrum Disorder (ASD) is a neurological and developmental condition that affects children’s social and cognitive skills, leading to repetitive behaviors, challenges in social interaction, communication difficulties, and restricted interests. Early diagnosis of autism can help mitigate its severity and long-term effects. This study proposes an automated electroencephalography (EEG)-based ASD detection method using data from Iraq and Poland. EEG signals underwent preprocessing, which included noise removal using a band-pass filter and artifact subspace reconstruction to ensure clean signals. Following preprocessing, features were extracted from the EEG channel power spectral density (PSD), and topographic brain maps (TBMs) were generated as inputs for deep feature extraction models, including AlexNet and GoogLeNet. Analysis of variance (ANOVA) was employed for feature selection (FS). Two types of linear classifiers, namely linear Support Vector Machine (SVM-L) and Linear Discriminant Analysis (LDA), were used for classification. The alpha band yielded the highest accuracy, reaching 98% (Iraq dataset, GoogLeNet + FS + SVM-L) and 96.5% (Poland dataset, AlexNet-FC6 + FS + SVM-L). The proposed method was evaluated against previous approaches using the same datasets, showing a substantial improvement in performance. Moreover, the study applied cross-dataset validation combined with feature fusion, achieving an average accuracy approximately 93%. These results highlight the need to evaluate the proposed approach on larger datasets to further ensure model generalization.