Artificial Intelligence-Powered MRI Segmentation for Uterine Fibroid Mapping: A Proof-of- Concept Study

Background Uterine fibroids are prevalent benign tumors that cause significant morbidity. Accurate diagnosis and personalized management remain challenging due to variability in imaging interpretation and complex clinical factors. Artificial intelligence (AI), particularly integrative models combining imaging, clinical, and pathological data, holds promise in revolutionizing diagnostic precision. Objective To develop and evaluate a proof-of-concept (POC), artificial intelligence (AI) powered diagnostic and integrative decision support system for uterine fibroid segmentation on Magnetic resonance Imaging (MRI) aimed at improving diagnostic accuracy, reproducibility and serving as a foundation for future decision-support tools. Methods In this retrospective analysis, multiparametric Magnetic resonance Imaging (MRI) and clinical data from 98 women with diagnosed uterine fibroids, obtained between April 2024 and July 2025 were analyzed. A U-Net convolutional neural network (CNN) was trained on 75% of the dataset, validated on 15%, and tested on 15%. Performance was assessed using Dice similarity coefficient (DSC), sensitivity, specificity and F1 Score. Segmentations and Manual annotations by two expert radiologists served as the reference standard. Imaging features were explored in combination with clinical parameters, primarily to assess feasibility and generate preliminary insights into integrative decision support. Results Our POC- AI model achieved a mean Dice Similarity Coefficient (DSC) of 0.92 , indicating excellent agreement between automated segmentation and expert annotations. This high DSC reflects accurate fibroid boundary delineation with consistent performance across fibroids of varying size and morphology. The model demonstrated a sensitivity of 0.91 and specificity of 0.94 , confirming robust lesion detection and reliable exclusion of non-fibroid tissue. Precision and recall values were balanced, yielding F1-score of 0.91 , which underscores the system’s ability to minimize both false positives and false negatives. Conclusion This proof-of-concept study demonstrates that AI-based MRI segmentation achieves high accuracy in uterine fibroid mapping, with excellent agreement with expert-defined ground truth annotations. Integration of such systems into gynecological practice may enhance diagnostic confidence, streamline surgical navigation, and lay the foundation for future tools supporting early malignancy detection and personalized treatment strategies.