A Novel EEG-Based Topographic Brain Map-Driven Deep Learning Method for Autism Spectrum Disorder Detection in...

Area of Science:

• Neuroscience
• Computational Psychiatry
• Medical Imaging

Background:

• Autism Spectrum Disorder (ASD) is a developmental condition impacting social and cognitive skills.
• Early ASD diagnosis is crucial for mitigating long-term effects and severity.
• Automated diagnostic tools can aid in timely and accurate identification.

Purpose of the Study:

• To develop and validate an automated electroencephalography (EEG)-based method for Autism Spectrum Disorder (ASD) detection.
• To assess the efficacy of deep learning models (AlexNet, GoogLeNet) combined with feature selection and linear classifiers for ASD classification.
• To evaluate the proposed method's performance on independent datasets from Iraq and Poland.

Main Methods:

• EEG signal preprocessing included band-pass filtering and artifact subspace reconstruction.
• Feature extraction involved power spectral density (PSD) and topographic brain maps (TBMs).
• Deep learning models (AlexNet, GoogLeNet), ANOVA for feature selection, and linear classifiers (SVM-L, LDA) were utilized.

Main Results:

• The alpha band demonstrated the highest classification accuracy: 98% for the Iraq dataset and 96.5% for the Poland dataset.
• The proposed method significantly outperformed previous approaches on the same datasets.
• Cross-dataset validation with feature fusion achieved an average accuracy of approximately 93%.

Conclusions:

• The developed automated EEG-based method shows high potential for accurate ASD detection.
• The findings underscore the effectiveness of deep learning and feature selection in EEG-based neurological disorder diagnosis.
• Further validation on larger, diverse datasets is recommended to ensure robust model generalization.