GlioSurvQNet: A DuelContextAttn DQN Framework for Brain Tumor Prognosis with Metaheuristic Optimization

Area of Science:

• Neuro-oncology
• Artificial Intelligence
• Medical Imaging

Background:

• Accurate brain tumor classification and survival prediction are crucial for neuro-oncology clinical decisions.
• Conventional methods face challenges like data scarcity, class imbalance, and poor generalization.
• GlioSurvQNet is introduced as a reinforcement learning framework to overcome these limitations.

Purpose of the Study:

• To develop and evaluate GlioSurvQNet, a novel reinforcement learning framework for glioma grading and survival prediction.
• To address limitations of existing machine learning and deep learning models in neuro-oncology.
• To enhance model interpretability and robustness in clinical settings.

Main Methods:

• GlioSurvQNet utilizes a DuelContextAttn Deep Q-Network (DQN) architecture for classification and survival prediction.
• Radiomics features were extracted from multimodal MRI (FLAIR, T1CE, T2) and optimized using metaheuristic algorithms (HHO, mGTO, ZOA).
• SHAP-based feature selection was employed for enhanced interpretability and robustness.

Main Results:

• The classification module achieved 99.27% accuracy with FLAIR + T1CE.
• The survival prediction model reached 93.82% accuracy with FLAIR + T2 + T1CE fusion.
• GlioSurvQNet outperformed established machine learning and deep learning models in both tasks.

Conclusions:

• GlioSurvQNet provides an accurate and interpretable AI-driven solution for brain tumor analysis.
• The framework demonstrates high accuracy and robustness, promising for clinical decision support.
• It aids in glioma diagnosis and prognosis, advancing precision neuro-oncology.