Adenomyosis tracking in MRI using deep learning

Pelvis, MR, Diagnostic procedure, Tissue characterisation Authors: L. F. D. Silva, D. L. D. Siqueira, E. D. S. Ribeiro, S. D. Castro, I. Coimbra Ladeira Morais, D. Monteiro Soares, S. Rego, A. P. Miranda Rosati, C. A. P. Fontes DOI: 10.26044/ecr2025/C-25955 Purpose Adenomyosis has a considerable negative impact on patients' quality of life. Despite being a benign gynecological disease, it causes various symptoms such as chronic pelvic pain, dysmenorrhea, dyspareunia, infertility, and abnormal uterine bleeding [1-12]. Previously, it was believed to primarily affect women over 40 years old; however, it is now identified in up to 30% of younger women and in 24% of those suffering from infertility based on imaging tests [3,4]. Studies have shown an association between adenomyosis and an increased risk of preterm birth...

and materials The study received approval from the responsible institutional ethics committee, ensuring compliance with ethical and regulatory guidelines. All patient information was carefully anonymized to protect data privacy and confidentiality. The study was registered in the Brazilian research platform under registration number CAAE: 56305321.9.0000.5259.A total of 598 pelvic MRI images were used, collected from 85 patients between June 2022 and August 2024, with an average of approximately seven uterine images per patient. The patient cohort represented diverse clinical conditions, including isolated adenomyosis, adenomyosis accompanied by myometrial...

To assess the performance of the model developed during neural network training, a held-out test set consisting of 120 MRI images was used. A comprehensive evaluation was conducted using various classification metrics including accuracy, precision, F1-score, sensitivity, specificity, area under the receiver operating characteristic (ROC) curve, and area under the precision-recall curve. The confusion matrix is presented in Table 3, and the quantitative results obtained for each metric are summarized in Table 4.Table 3 (confusion matrix) presents the classification result by the neural network on...

This study introduces an efficient model for screening adenomyosis in uterine magnetic resonance imaging, that, to the best of our knowledge, has not been previously employed for this purpose. The results are promising, although the study has limitations such as training on a reduced dataset. Future studies will involve testing additional convolutional neural network architectures and expanding the image dataset. This approach reinforces the feasibility of applying artificial intelligence in medical contexts and opens new avenues for future research to enhance applicability and accuracy. Personal information and conflict of interest L. F. D. Silva: Nothing to disclose D. L. D. Siqueira: Nothing to disclose E. D. S. Ribeiro: Nothing to disclose S. D. Castro: Nothing to disclose I. Coimbra Ladeira Morais: Nothing to disclose D. Monteiro Soares: Nothing to disclose S. Rego: Nothing to disclose A. P. Miranda Rosati: Nothing to disclose C. A. P. Fontes: Nothing to disclose

Upson K, Missmer SA. Epidemiology of adenomyosis. Semin Reprod Med. 2020;38(2):89-107. doi: 10.1055/s-0040-1718920. Gordts S, Grimbizis G, Campo R. Symptoms and classification of uterine adenomyosis, including the place of hysteroscopy in diagnosis. Fertil Steril. 2018;109(3):380-388.e1. doi: 10.1016/j.fertnstert.2018.01.006. Abu Hashim H, Elaraby S, Fouda AA, Rakhawy ME. The prevalence of adenomyosis in an infertile population: a cross-sectional study. Reprod Biomed Online. 2020;40(6):842-850. doi: 10.1016/j.rbmo.2020.02.011. Puente JM, Fabris A, Patel J, Patel A, Cerrillo M, Requena A, et al. Adenomyosis in infertile women: prevalence and the role...