Convolutional neural network framework for EEG-based ADHD diagnosis in children.

Area of Science:

• Neuroscience
• Computational Psychiatry
• Medical Imaging

Background:

• Attention-deficit hyperactivity disorder (ADHD) is a prevalent neuro-developmental disorder affecting school children globally.
• Prevalence rates in India range from 5% to 11%, necessitating effective diagnostic tools.
• Electroencephalography (EEG) offers a promising avenue for early ADHD detection and classification.

Purpose of the Study:

• To introduce a simple yet effective Convolutional Neural Network (CNN) architecture for ADHD detection in children.
• To evaluate the performance of the CNN using pre-processed EEG signals.
• To explore the feasibility of using a reduced set of EEG channels for diagnosis.

Main Methods:

• EEG signals were pre-processed using a band-pass filter and segmented into 5-second frames.
• Frames underwent normalization and canonical correlation analysis before CNN application.
• A two-convolutional-layer CNN architecture was employed for training and testing.

Main Results:

• 100% accuracy, sensitivity, and specificity were achieved using all 19 EEG channels.
• Investigated using only frontal EEG channels for reduced computational complexity.
• Frontal channel analysis yielded an accuracy of 99.08% for ADHD detection.

Conclusions:

• The proposed CNN method demonstrates high accuracy and ease of implementation for ADHD diagnosis.
• The approach holds significant potential for widespread clinical deployment.
• Simplified EEG channel usage offers a practical balance between accuracy and computational efficiency.