{"paper_id":"2ae7da64-7508-4162-9ff3-21489b5b2a53","body_text":"ECR 2025 / C-14859\nDevelopment of a deep learning algorithm for the detection of endometriosis lesions in pelvic MRI\nCongress:\nECR 2025\nPoster Number:\nC-14859\nType:\nScientific Exhibit\nKeywords:\nArtificial Intelligence, Genital / Reproductive system female, Pelvis, MR, Neural networks, Diagnostic procedure, Tissue characterisation\nAuthors:\nM. Florin, V. Muñoz-Ramírez, S. Berthoumieux, K. Mignon-Godefroy, S. Doyle, C. Real, L. Jarboui\nDOI:\n10.26044/ecr2025/C-14859\nPurpose\nEndometriosis is a chronic condition affecting approximately 10% of reproductive-age women, characterized by the presence of endometrial-like tissue outside the uterus. This ectopic tissue can lead to significant pain, infertility, and reduced quality of life. Accurate diagnosis is essential for patient management, yet it remains challenging due to the variability in lesion presentation and the expertise required for interpretation. This study presents SMART-IRM, an artificial intelligence (AI) algorithm designed to assist in the detection of endometriosis lesions on pelvic MRI scans. The software aims to...\nMethods and materials\nMedical imaging plays a crucial role in diagnosing endometriosis and planning surgical interventions. MRI has been shown to be highly accurate, with similar performance to ultrasound for detecting ovarian endometriomas and superior sensitivity for deep pelvic infiltrating endometriosis (Nisenblat, 2016; Bazot, 2009). However, MRI-based lesion detection depends on radiologist experience, with reported sensitivity for deep lesions ranging from 80% for experienced radiologists to as low as 40% for less experienced practitioners (Saba, 2011). To address this challenge, we developed SMART-IRM, a deep learning model trained...\nResults\nThe algorithm was evaluated on a validation set comprising 52 patients imaged at two medical centers using five different MRI machines. SMART-IRM achieved a sensitivity of 76% in detecting deep infiltrating endometriosis when compared to expert-validated annotations.To assess human interrater variability, two radiologists independently reviewed a subset of 37 patients from the validation cohort, achieving an interobserver sensitivity of 85%.\nConclusion\nThese preliminary results demonstrate the feasibility of using AI to support the diagnosis of endometriosis in pelvic MRI scans. SMART-IRM shows promise as a diagnostic support tool, helping to bridge performance gaps between expert and non-expert radiologists. Future work will focus on expanding the model’s capabilities to include a wider range of anatomical sites and more complex cases, such as bowel and bladder involvement. Additionally, the model’s ability to identify associated findings such as adenomyosis and hematosalpinx will be further explored, potentially improving comprehensive endometriosis...\nPersonal information and conflict of interest\nM. Florin:\nNothing to disclose\nV. Muñoz-Ramírez:\nEmployee: Pixyl\nS. Berthoumieux:\nEmployee: Endodiag\nK. Mignon-Godefroy:\nEmployee: Endodiag\nS. Doyle:\nCEO: Pixyl\nC. Real:\nCEO: Medevice\nCEO: Endodiag\nL. Jarboui:\nNothing to disclose\nReferences\nThis study is limited by its sample size and focus on specific lesion types. While the model demonstrated promising sensitivity, additional work is required to optimize performance, particularly for less common lesion location. Further validation with larger, more diverse datasets is necessary to confirm the model’s generalizability across different imaging protocols and patient populations. Additionally, while AI can support radiologists, its clinical integration and impact on workflow efficiency remain to be fully assessed in real-world settings.","source_license":"CC0","license_restricted":false}