Classification of attention deficit/hyperactivity disorder based on EEG signals using a EEG-Transformer model^∗

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Diagnostics

Background:

• Attention-Deficit/Hyperactivity Disorder (ADHD) is a prevalent neurodevelopmental disorder in adolescents.
• Current ADHD diagnosis relies on subjective assessments, leading to potential delays and misdiagnoses.
• Objective diagnostic methods are needed to improve ADHD detection efficiency.

Purpose of the Study:

• To develop and evaluate a deep learning model for objective ADHD diagnosis using electroencephalogram (EEG) signals.
• To compare the performance of the proposed EEG-Transformer model against existing convolutional neural network (CNN) models.
• To establish an auxiliary diagnostic tool for clinicians and a foundation for transferable EEG learning.

Main Methods:

• Proposed the EEG-Transformer deep learning model, leveraging attention mechanisms for EEG signal feature extraction and classification.
• Conducted a comparative analysis against three established CNN models.
• Performed ablation and optimization experiments to validate model components and performance.

Main Results:

• The EEG-Transformer model achieved an average accuracy of 95.85% and an AUC of 0.9926.
• Demonstrated superior performance and faster convergence compared to the evaluated CNN models.
• Ablation studies confirmed the functional significance of model modules.

Conclusions:

• The EEG-Transformer model shows significant potential as an objective, accurate auxiliary tool for clinical ADHD diagnosis.
• This model provides a robust baseline for future research in transferable learning for EEG signal classification.
• The findings suggest a promising direction for improving diagnostic efficiency and accuracy in neurodevelopmental disorders.