Thyroid disease classification using generative adversarial networks and Kolmogorov-Arnold network for three-class...

Area of Science:

• Artificial Intelligence in Medicine
• Machine Learning for Diagnostics
• Medical Data Augmentation

Background:

• Accurate thyroid disease classification (hyperthyroidism, hypothyroidism, normal) is crucial for medical diagnostics.
• Traditional diagnostic methods face challenges with class imbalance and limited data.
• Need for advanced AI techniques to improve accuracy and interpretability in thyroid disease diagnosis.

Purpose of the Study:

• To develop and evaluate an advanced machine learning approach for thyroid disease classification.
• To investigate the efficacy of generative adversarial networks (GANs) for data augmentation in this context.
• To assess the performance of Kolmogorov-Arnold networks (KANs) and compare them with traditional models.

Main Methods:

• Integration of generative adversarial networks (GANs) for synthetic data generation and augmentation.
• Implementation and evaluation of Kolmogorov-Arnold networks (KANs) for thyroid disease classification.
• Comparison with traditional machine learning models (logistic regression, random forest, SVM, MLP) and use of SHAP/LIME for interpretability.

Main Results:

• GAN-based data augmentation significantly improved classification accuracy, especially for minority classes.
• The Kolmogorov-Arnold network (KAN) model achieved 98.68% accuracy, outperforming traditional neural networks.
• Thyroid stimulating hormone identified as the most prominent predictive feature using explainability methods (SHAP, LIME).

Conclusions:

• The combination of GANs and KANs offers a robust and accurate framework for thyroid disease classification.
• This AI-driven approach effectively addresses class imbalance and enhances model generalization capabilities.
• The developed explainable AI models provide a feasible foundation for clinical decision support systems, improving patient outcomes.