IMPROVING PREDICTION ACCURACY OF DEEP LEARNING FOR BRAIN CANCER DIAGNOSIS USING POLYAK-RUPPERT OPTIMIZATION

Accurate and reliable diagnosis is critical for effective treatment planning for brain cancer. Recent advancements in deep learning have significantly enhanced diagnostic capabilities, but challenges persist in optimizing model performance for diverse and complex datasets. This study investigates the application of Polyak-Ruppert Optimization (PRO) to improve the prediction accuracy of conventional deep learning models for brain cancer diagnosis. Utilizing the REpository of Molecular BRAin Neoplasia DaTa (REMBRANDT) database, the proposed framework incorporates the advanced PRO technique to stabilize training and enhance generalization. The PRO’s impacts on convergence rates, model robustness, and predictive accuracy across multiple cancer types are analyzed. Experimental results demonstrate that VGG and ResNet models employing the PRO technique outperform the conventional architectures such as VGG and ResNet in classification metrics such as accuracy, sensitivity, and specificity. The potential of advanced optimization strategies such as PRO to refine deep learning applications in oncology paves the way for more accurate, efficient, and interpretable diagnostic systems.