Design and Development of Meta-Heuristic Transfer Learning ‎Techniques for Ovarian Cancer Detection

Cancer continues to claim a significant number of lives globally, with ovarian cancer being particularly deadly owing to its late-stage ‎diagnosis and limited availability of effective treatments. Current diagnostic methods often rely on traditional imaging and histopathological ‎analysis, which face challenges such as limited accuracy, high false-positive rates, and dependency on expert interpretation. Though artificial ‎intelligence techniques have addressed this bottleneck, achieving higher performance with low computational overhead remains a significant ‎challenge in current research. To rectify these drawbacks, this research recommends a hybrid approach leveraging transfer learning models ‎such as Inception V3 and Improved Black Widow Optimized dense classification layers to enhance diagnosis performance. The model was ‎trained and validated using the UBC-OCEAN dataset, which consists of 5,634 standardized samples classified into three primary ovarian ‎cancer subtypes: high-grade serous carcinoma (3,412 samples), endometrioid carcinoma (1,732 samples), and clear cell carcinoma (490 ‎samples), alongside rare outlier cases for anomaly detection. The comprehensive outcomes portray the recommended approach attained the ‎classification accuracy of 98.7%, with a precision of 98.7%, recall of 98.7%, and an F1-score of 98.7%, demonstrating enhanced ‎performance compared to conventional Deep Learning (DL) approaches. The integration of metaheuristic optimization further refined the ‎classification process, enhancing model performances and robustness. The results emphasize the model’s efficacy as a reliable tool for early ‎detection and subtype identification, and deployment in clinical workflows, offering scalability and ease of integration into existing ‎diagnostic pipelines, ultimately contributing to improved patient diagnosis outcomes‎.