Artificial intelligence applications in ovarian cancer detection: A systematic literature review of deep learning approaches and clinical translation challenges
This systematic literature review examines artificial intelligence applications in ovarian cancer detection through analysis of 61 studies published between 2020 and 2025. The investigation reveals advancing methodological sophistication alongside persistent limitations constraining clinical translation. Convolutional neural network architectures dominate current research, with ResNet variants achieving accuracy rates between 92 % and 99.7 %. Vision transformer integration demonstrates competitive performance while providing attention-based interpretability mechanisms. Ensemble methodologies produce superior diagnostic accuracy, reaching 98.96 % through multi-model integration strategies. Dataset heterogeneity presents barriers to model generalizability, with sample sizes ranging from hundreds to thousands of images and private institutional datasets limiting external validation opportunities. Hyperparameter optimization receives minimal attention across reviewed studies, with twelve investigations implementing systematic parameter tuning. Explainable AI implementation occurs in seven studies despite growing regulatory requirements for transparent medical AI systems. Object detection applications demonstrate limited adoption compared to classification approaches, with U-Net variants comprising the primary segmentation methodology. Bibliometric analysis indicates increasing research activity, with publication counts rising from 5 papers in 2020-2021, 34 papers in 2023-2024. Future research directions should prioritize standardization initiatives for data collection and evaluation metrics, prospective clinical trials for validation, and regulatory pathway development for medical AI system approval.
