Proposing an adaptive mutation to improve XCSF performance to classify ADHD and BMD patients

Area of Science:

• Neuroscience
• Computational Psychiatry
• Biomedical Engineering

Background:

• Clinical symptoms of bipolar mood disorder (BMD) and attention deficit hyperactivity disorder (ADHD) in children significantly overlap, complicating accurate diagnosis.
• Current diagnostic methods relying on clinical symptoms lack precision, necessitating quantitative criteria for objective differentiation.

Purpose of the Study:

• To design an efficient decision-making system for accurate classification of ADHD and BMD patients using electroencephalogram (EEG) signals.
• To develop a robust and accurate method for discriminating between pediatric ADHD and BMD.

Main Methods:

• Recorded 22-channel EEGs from 21 ADHD and 22 BMD subjects.
• Extracted features including fractal dimension, band power, and autoregressive coefficients.
• Utilized linear discriminant analysis (LDA) for dimensionality reduction and a modified extended classifier system for function approximation (XCSF) with an adaptive mutation rate for classification.

Main Results:

• The proposed XCSF classifier demonstrated robust performance in discriminating between ADHD and BMD patients, even in simulated noisy environments.
• Compared to conventional classifiers like support vector machine (SVM), LDA, and nearest neighbor, the proposed method showed superior robustness.
• Statistical tests confirmed the proposed classifier's effectiveness in differentiating between the two disorders.

Conclusions:

• The developed EEG-based quantitative approach offers a promising tool for objective discrimination between pediatric ADHD and BMD.
• The adaptive mutation rate in the XCSF classifier enhances classification accuracy and avoids premature convergence.
• This technique can aid clinicians in making more precise diagnoses, leading to improved treatment outcomes for children.