Novel explainable transformer model for automated attention-deficit hyperactivity disorder detection with EEG signals

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Diagnostics

Background:

• Deep learning advances drive interest in medical applications like automated neurodevelopmental disorder detection.
• Attention deficit hyperactivity disorder (ADHD) is a prevalent neurodevelopmental condition often diagnosed in childhood.
• Electroencephalogram (EEG) signals are commonly used for ADHD detection but are nonlinear, non-stationary, and variable, posing challenges for automated methods.

Purpose of the Study:

• To develop an explainable artificial intelligence (AI) model for automated ADHD detection from EEG signals.
• To enhance trust and transparency in AI-based medical diagnostic tools.
• To provide clinicians with an interpretable AI system for ADHD screening.

Main Methods:

• Proposed an explainable transformer-based model for ADHD detection using EEG signals.
• Employed Local Interpretable Model-agnostic Explanations (LIME) for visualizing model predictions.
• Utilized specific EEG signal components: delta (1-3 Hz), theta (4-7 Hz), and beta (14-30 Hz) for analysis.

Main Results:

• Achieved an average accuracy of 96.2% using leave-one-out cross-validation on a public dataset.
• Demonstrated that the model's explanations are clinically meaningful and align with accepted ADHD markers.
• Validated the model's effectiveness in classifying ADHD from EEG data.

Conclusions:

• The explainable transformer model is successful for ADHD detection.
• The model has the potential to increase clinician trust in AI tools for ADHD screening.
• This approach can serve as an auxiliary tool in healthcare settings for ADHD assessment.