Wavelet-Attention deep model for pediatric ADHD diagnosis via EEG

Area of Science:

• Neuroscience
• Biomedical Engineering
• Artificial Intelligence

Background:

• Attention-Deficit/Hyperactivity Disorder (ADHD) is a common childhood neurodevelopmental disorder.
• Current ADHD diagnostic methods often rely on subjective assessments, highlighting the need for objective tools.
• Early diagnosis is critical for effective intervention and improving outcomes.

Purpose of the Study:

• To develop and validate a novel Wavelet-Attention deep model for objective ADHD diagnosis using electroencephalography (EEG) signals.
• To enhance the accuracy and objectivity of ADHD diagnosis in children.
• To identify key neurophysiological markers associated with ADHD through model interpretability.

Main Methods:

• Utilized a publicly available EEG dataset from 121 children.
• Applied Independent Component Analysis (ICA) for artifact removal and preprocessing.
• Developed a deep learning model integrating wavelet transform for feature extraction and a ResNet with an attention mechanism.
• Employed Leave-One-Subject-Out cross-validation for robust evaluation.

Main Results:

• The Wavelet-Attention deep model achieved high diagnostic performance: 96.69% accuracy, 95.08% sensitivity, and 98.33% specificity.
• Model interpretability indicated that frontal lobe EEG channels and low-frequency wavelet features are crucial for ADHD identification.
• Results align with known neurophysiological characteristics of ADHD.

Conclusions:

• The proposed model demonstrates significant potential as a reliable and objective tool for ADHD diagnosis.
• This objective approach can facilitate earlier detection and personalized interventions for children with ADHD.
• Further development could lead to clinical integration for improved ADHD management.