Siamese based deep neural network for ADHD detection using EEG signal.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Pediatric Medicine

Background:

• Early detection of Attention-Deficit/Hyperactivity Disorder (ADHD) in children is vital for effective intervention and tailored treatment.
• Understanding the neurobiological underpinnings of ADHD is essential for improving diagnostic accuracy.

Purpose of the Study:

• To apply deep learning models to analyze Electroencephalography (EEG) derived brain maps for ADHD detection in pediatric subjects.
• To leverage explainable AI (XAI) to identify key brain regions and signal features indicative of ADHD.

Main Methods:

• A Siamese-based Convolutional Neural Network (CNN) was utilized to process EEG-based brain maps.
• Power Spectral Density (PSD) analysis was performed on EEG signals.
• Gradient-weighted Class Activation Mapping (Grad-CAM) was employed for feature visualization and interpretation.

Main Results:

• The CNN model achieved a high classification accuracy of 99.17% for ADHD detection.
• Grad-CAM analysis identified theta band PSD features from the frontal and occipital lobes as significant discriminators.
• These findings highlight specific neurophysiological markers associated with ADHD in children.

Conclusions:

• Deep learning, particularly CNNs, demonstrates high efficacy in detecting ADHD in pediatric populations.
• Regional PSD metrics, especially in the theta band of frontal and occipital lobes, are crucial for accurate ADHD classification.
• Explainable AI (Grad-CAM) enhances the understanding of ADHD neurobiology, paving the way for improved diagnostic precision.