EEG-Based ADHD Classification Using Autoencoder Feature Extraction and ResNet with Double Augmented Attention...

Area of Science:

• Neuroscience
• Medical Technology
• Artificial Intelligence

Background:

• Attention-Deficit/Hyperactivity Disorder (ADHD) is a common neurodevelopmental disorder with no objective diagnostic tests.
• Current ADHD diagnosis relies on subjective psychiatric assessments, necessitating significant clinical effort.
• There is a need for objective tools to enhance ADHD diagnostic accuracy and reduce subjectivity.

Purpose of the Study:

• To develop an objective diagnostic method for ADHD using electroencephalography (EEG) signal analysis.
• To leverage deep learning techniques for improved ADHD diagnosis from EEG data.
• To introduce a novel ResNet-based model with a double-augmented attention mechanism for complex EEG pattern recognition.

Main Methods:

• Utilized an autoencoder for feature extraction from EEG data.
• Employed the Reptile Search Algorithm for optimal feature selection.
• Developed and trained a modified ResNet architecture for ADHD classification.

Main Results:

• The proposed ResNet model demonstrated superior performance compared to traditional classifiers.
• Achieved 99.42% accuracy, 99.03% precision, 99.82% recall, and 99.42% F1-score.
• The model achieved a Receiver Operating Characteristic Area Under the Curve (ROC AUC) score of 0.99.

Conclusions:

• The ResNet model shows exceptional capability in differentiating between children with and without ADHD.
• The objective diagnostic approach minimizes misclassification errors.
• This advanced EEG analysis offers improved diagnostic precision for ADHD.