Abstract
Category: AI education and research: examples of proof of concept or AI in development, technical advances, teaching approaches or preclinical testingUterine fibroids are common benign tumours that contribute substantially to morbidity.Their evaluation and management remain complex, largely due to heterogeneity in imaging interpretation and clinical decision-making.Artificial intelligence (AI) models that integrate imaging, clinical and pathological information have the potential to transform diagnostic accuracy and reproducibility.Our objective is to design and assess a proof of concept (POC) AI-driven approach for uterine fibroid segmentation on magnetic resonance imaging (MRI), with the goal of enhancing diagnostic precision and consistency, and establishing a basis for future decision-support applications.This retrospective study analysed multiparametric MRI scans and clinical data from 98 women with confirmed fibroids collected between January 2023 and February 2025.A U-Net convolutional neural network (CNN) was trained on 75% of cases, with 15% used for validation and 15% for testing.Model performance was quantified using the Dice similarity coefficient (DSC), sensitivity, specificity and F1 score.Segmentations generated by the AI were benchmarked against manual annotations from two expert radiologists.Imaging outputs were also examined alongside clinical data to assess the feasibility of integrated decision support.The AI model achieved a mean DSC of 0.92, showing close alignment with expert annotations and accurate fibroid boundary delineation across diverse lesion sizes and morphologies.Sensitivity was 0.91 and specificity 0.94, reflecting strong lesion detection and low misclassification of non-fibroid tissue.Balanced precision and recall produced an overall F1 score of 0.91, confirming reliable system performance.This POC work demonstrates that AI-based MRI segmentation can deliver high accuracy in uterine fibroid mapping, closely mirroring expert assessment.Incorporation of such approaches into gynaecological imaging practice could enhance diagnostic certainty, assist surgical planning and provide a platform for future tools aimed at early malignancy detection and personalised treatment pathways.