Artificial intelligence as a diagnostic support tool in hysteroscopy: current evidence and clinical implications.
Authors
Affiliations (10)
Affiliations (10)
- Department of Obstetrics and Gynecology, Gynecologic Oncology and Minimally Invasive Pelvic Surgery, International School of Surgical Anatomy (ISSA), IRCCS Sacro Cuore - Don Calabria Hospital, Negrar di Valpolicella, Verona, Italy.
- Gynecologic Oncology Unit, Fondazione IRCCS Istituto Nazionale dei Tumori, Milan, Italy.
- Gynecologic Oncology Unit, Department of Women's and Children's Health and Public Health, Fondazione Policlinico Universitario A. Gemelli, IRCCS, Rome, Italy.
- IHU Strasbourg, Institute of Image-Guided Surgery, Strasbourg, France.
- IRCAD, Research Institute against Digestive Cancer (IRCAD) France, Strasbourg, France.
- Obstetrics and Gynecology Unit, Maternal and Child Department, San Pietro Fatebenefratelli Hospital, Rome, Italy.
- Department of Translational Medicine, University of Ferrara, Ferrara, Italy.
- Department of Surgical Gynecology, University of Clermont Auvergne, Clermont-Ferrand, France.
- Department of Gynaecological Oncology, Churchill Cancer Centre, Oxford University Hospitals NHS Foundation Trust, Oxford, UK.
- Department of Clinical and Experimental Sciences, University of Brescia, Brescia, Italy.
Abstract
Hysteroscopy allows direct inspection of the uterine cavity for many conditions. Despite being widely adopted, its diagnostic accuracy largely depends on surgeon expertise, leading to potentially misleading diagnoses. Artificial intelligence (AI) has shown robust performance in many areas of medical imaging. The application of AI to hysteroscopy can improve diagnostic reliability and clinical decision-making. We carried out a systematic review following PRISMA guidelines to summarize current evidence on the use of AI in hysteroscopy. Eligible studies involved human subjects undergoing hysteroscopy in which AI models were applied to image or video data for diagnostic, classification, or prognostic purposes. Literature searches were conducted in PubMed, Scopus, and Web of Science up to August 2025. We extracted details on design, patient population, AI architecture, dataset characteristics, validation approach, and performance outcomes. Study quality and risk of bias were assessed with the Newcastle-Ottawa Scale. Fifteen studies published between 2021 and 2025 met the inclusion criteria. Applications of AI in hysteroscopy clustered around three major domains: intrauterine adhesions (IUAs), chronic endometritis (CE), and intracavitary lesions. For IUAs, predictive models demonstrated strong performance, with AUC values up to 0.99 for fertility and recurrence outcomes, highlighting the potential for AI to support tailored postoperative care. For CE, both Convolutional Neural Network-based methods and spectroscopy-assisted approaches achieved diagnostic concordance with histopathology exceeding 80-90%, suggesting that AI could potentially reduce the need for biopsy. In lesion classification, models achieved accuracies above 85% and, in some cases, outperformed gynecologists in distinguishing benign from malignant findings. Early work on real-time video analysis also demonstrated promise for intraoperative support. Overall, the quality of the studies included was moderate to high. AI applied to hysteroscopy shows considerable promise for enhancing diagnostic accuracy, consistency, and intraoperative decision-making. To enable translation into practice, future research should emphasize multicenter collaborations, standardized imaging protocols, external validation, and the development of explainable models that can be trusted in clinical settings.