Deep Learning-Based Automated Segmentation of Uterine Myomas

achieved a mean Dice score of 0.70 for this class (Table I), highlighting its robustness in detecting and delineat- ing myomas across diverse cases. The large standard deviation reflects substantial case-to-case variability, as also visualized in Figure 3 where some cases exhibit Dice scores below 0.50, while some others exceed 0.80. This suggests that while the model can segment myomas accurately in favorable cases, challenges remain for cases with atypical morphology or low contrast. Such variability underscores the importance of further refinements, such as tailored post-processing or additional training data focused on difficult cases. Beyond myoma segmentation, the proposed method also performed impressively across other classes. It achieved Dice scores of 0.86 for the uterine wall, 0.79 for the uterine cavity, and 0.68 for the nabothian cyst (Table I), demonstrating the method’s ability to generalize across multiple classes. Notably, uterine wall segmentation was highly consistent across cases, as reflected by its low standard deviation and compact box plot distribution (Table I, Figure 3). In contrast, the lower and more variable scores for the nabothian cyst reflect expected challenges associated with its small size. Overall, the pro- posed method achieved a mean Dice score of 0.76 across all classes, validating its effectiveness as a powerful framework for uterine MRI segmentation. Crucially, its performance on myoma segmentation establishes a significant step forward in the development of automated tools for managing one of the most prevalent and burdensome conditions in women’s reproductive health. TABLE I DICE SCORES (MEAN ± STANDARD DEVIATION ) ON THE TEST SET . Label Dice Score Uterine Wall 0.86 ± 0.05 Uterine Cavity 0.79 ± 0.10 Myoma 0.70 ± 0.27 Nabothian Cyst 0.68 ± 0.38 Fig. 3. Dice Scores for uterine wall, uterine cavity, myoma and nabothian cyst on the test set. IV. C ONCLUSIONS In this study, we addressed the need for an automated and reproducible method for uterine myoma segmentation by leveraging a publicly available MRI dataset. The results demonstrate that our proposed method can reliably segment uterine myomas along with surrounding structures, offering a robust baseline for future research. This approach holds promise for reducing the clinical burden associated with manual segmentation and enabling more standardizeded as- sessment of myoma characteristics.

[1] M. M. McWilliams and V . M. Chennathukuzhi, “Recent advances in uterine fibroid etiology,” in Seminars in reproductive medicine , vol. 35, no. 02. Thieme Medical Publishers, 2017, pp. 181–189. [2] A. Tinelli, A. Morciano, R. Sparic, S. Hatirnaz, L. E. Malgieri, A. Mal- vasi, A. D’Amato, G. M. Baldini, and G. Pecorella, “Artificial intelligence and uterine fibroids: A useful combination for diagnosis and treatment,” Journal of Clinical Medicine , vol. 14, no. 10, p. 3454, 2025. [3] H. Pan, M. Chen, W. Bai, B. Li, X. Zhao, M. Zhang, D. Zhang, Y . Li, H. Wang, H. Genget al., “Large-scale uterine myoma mri dataset covering all figo types with pixel-level annotations,” Scientific Data, vol. 11, no. 1, p. 410, 2024. [4] F. Isensee, P. F. Jaeger, S. A. Kohl, J. Petersen, and K. H. Maier-Hein, “nnu-net: a self-configuring method for deep learning-based biomedical image segmentation,” Nature methods, vol. 18, no. 2, pp. 203–211, 2021.