Explainability, Uncertainty, and Clinical Trust in Artificial Intelligence for Endometriosis: A Scoping Review

Endometriosis is a common, chronic gynaecological condition that remains difficult to diagnose, with delays of 7 to 10 years between symptom onset and confirmed diagnosis. Definitive diagnosis often requires invasive laparoscopy, and the non-invasive imaging alternatives demand subspecialist expertise that is not widely available. Artificial intelligence (AI) and machine learning (ML) have been proposed as tools to support earlier and more accurate diagnosis, with rapid growth in the literature since 2020. Clinical diagnosis typically draws on several sources of information together, including symptoms, imaging, and laboratory markers. Models that combine these data types are therefore expected to outperform single-modality approaches by exploiting complementary signals, which makes the extent and success of multimodal integration a question of particular interest. However, acceptability of AI use in clinical settings depends not only on predictive performance but also on explainability, calibration, uncertainty quantification, and consideration of ethical, governance, and trust-related factors. Whether the published literature reports these features has not been systematically examined. Adenomyosis, a related uterine condition that frequently co-occurs with endometriosis and shares diagnostic challenges, was also included in the registered search scope.