Artificial Intelligence for Ovarian Cancer Detection with Medical Images: A Review of the Last Decade (2013–2023)

Abstract

The symptoms of ovarian cancer are nonspecific, and current screening methods lack sufficient accuracy for early diagnosis. This often leads to detection at a later, more advanced stage of the disease. Medical imaging provides morphological and functional data to help characterize ovarian tumors, but more research is needed to develop reliable early screening tools. This review examines recent machine learning techniques applied to imaging data for improving ovarian cancer detection and diagnosis. A literature search was conducted on PubMed, IEEE, and ACM databases for studies from 2010 to 2023 utilizing machine learning with ultrasound, magnetic resonance imaging, computed tomography, or other imaging data and clinical records to detect ovarian cancer. Key information extracted included imaging modality and clinical recordings, machine learning methods, classification tasks, performance metrics, and datasets. This work identified 81 relevant studies. Artificial intelligence approaches included traditional methods like support vector machines, random forest and logistic regression, and deep learning models like convolutional neural networks, vision transformers, and graph neural networks. Most studies focused on the binary classification of benign vs. malignant adnexal masses. The range of reported diagnostic accuracy across different modalities is 75–99%. Deep learning generally outperformed traditional machine learning models. Consequently, machine learning, especially deep learning, shows promising performance in detecting ovarian cancer from medical images. However, the heterogeneity of imaging protocols, data labeling biases, model interpretability, and validation on multi-center datasets is challenging. Future work should focus on robust and generalizable solutions that can be deployed as clinical tools for improving ovarian cancer outcomes.