Multi-class brain tumor diagnosis using MRI: A dynamic reinforcement ensemble learning approach.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Medical Informatics

Background:

• Accurate medical diagnosis is crucial for clinical decisions, especially with complex, imbalanced datasets.
• Traditional static ensemble models exhibit limitations in adapting to such data, impacting generalization and performance.

Purpose of the Study:

• To develop a novel dynamic ensemble learning framework for enhanced diagnostic accuracy.
• To address static ensemble limitations via adaptive model selection and dynamic weight adjustment.

Main Methods:

• A Dynamic Reinforcement Ensemble Learning Model utilizing reinforcement learning (RL) was proposed.
• The model dynamically selects base classifiers and adjusts their weights based on data characteristics.
• Evaluated on LSBTDK-DAT, FIGSHARE-DAT, and THYROID-DAT benchmark datasets with comparative and ablation studies.

Main Results:

• Achieved 99.55% accuracy and 99.54% F1-score on LSBTDK-DAT; 99.35% accuracy and 98.89% F1-score on THYROID-DAT.
• Outperformed state-of-the-art methods by up to 7% in accuracy and 5% in F1-score across datasets.
• Ablation studies confirmed superior performance from combined dynamic selection and weighting.

Conclusions:

• Dynamic reinforcement ensemble selection with adaptive weighting robustly handles complex medical data.
• Demonstrates potential for intelligent clinical decision support systems.
• Lays foundation for scalable, high-precision medical AI solutions.