Recent Advances in Semantic Segmentation and Classification for Ovarian Cancer Detection Using EfficientNetB0 with FPN and Causal Dilated Convolutional Neural Networks: A Systematic Review

Abstract

Ovarian cancer remains a leading cause of mortality among gynecological malignancies due to delayed diagnosis and the complexity of accurately identifying tumors in medical images. Recent advances in deep learning have significantly improved automated detection by enabling precise semantic segmentation and classification of ovarian lesions. Architectures such as EfficientNetB0 provide efficient feature extraction with reduced computational complexity, while Feature Pyramid Networks (FPN) enhance multi-scale feature representation for detecting tumors of varying sizes. Additionally, causal dilated convolutional neural networks improve contextual feature learning, leading to better segmentation accuracy and more reliable classification of benign and malignant tumors. The reviewed studies demonstrate that integrating EfficientNetB0, FPN, and advanced segmentation models such as U-Net substantially enhances diagnostic performance compared with conventional machine learning methods. Hybrid deep learning architectures and attention mechanisms further improve feature discrimination, enabling classification accuracies exceeding 90% in several studies. Nevertheless, challenges including limited annotated datasets, model generalization across diverse clinical environments, computational complexity, and limited interpretability continue to restrict large-scale clinical adoption. Future research should emphasize multimodal medical image integration, explainable artificial intelligence, lightweight deep learning models, and robust validation on diverse datasets. These advancements will support accurate, efficient, and clinically reliable ovarian cancer detection systems, ultimately improving early diagnosis, treatment planning, and patient outcomes.