Researchers at Adelaide University have developed an AI tool that identifies signs of endometriosis from a single MRI or ultrasound scan. The system, called EndoFusion, accurately detects two major imaging markers of advanced endometriosis during a standard pelvic examination. This development could reduce the reliance on invasive surgical confirmation for patients.
How the AI Scans Work
EndoFusion is designed to find specific signs on MRI and transvaginal ultrasound (TVUS) scans. The framework looks for pouch of Douglas (POD) obliteration and bowel nodules. It was trained on data from four datasets containing over 9,000 female pelvic MRI scans and more than 800 TVUS sliding scans.
Unlike previous systems, EndoFusion does not require both imaging types to be available for the same patient. This flexibility allows the AI to analyze either MRI or TVUS data independently. The system processes the images rapidly, detecting relevant signs in 18 milliseconds.
According to Associate Professor Jodie Avery, a co-author of the study and co-lead of the Endometriosis Research Group at Adelaide University’s Robinson Research Institute, the model achieved an average Area Under the Curve (AUC) of 0.827 across four imaging tasks. The specific results include an AUC of 0.756 for POD obliteration on MRI, 0.903 for bowel nodules on MRI, 0.907 for POD obliteration on TVUS, and 0.741 for bowel nodules on TVUS.
Performance and Clinical Use
Lead author Dr Yuan Zhang, also of Adelaide University’s Robinson Research Institute, noted that the model provided a correct diagnosis 83 percent of the time. This accuracy exceeds that of all competing models. The researchers tested the system at a threshold with a specificity of at least 90 percent to simulate a clinical environment where false-positive findings must be kept low.
At that specific operating point, the model showed varying sensitivity across different imaging tasks. Sensitivity reached 54.5 percent for bowel nodules on MRI and 79.6 percent for POD obliteration on TVUS.
Broader Impact and Future Steps
Dr Zhang added that the tool could also inform research into other conditions that use multimodal imaging, including other gynecological disorders, prostate and breast cancers, and fetal abnormalities. The researchers plan to expand EndoFusion to assess a broader range of findings routinely considered in specialist endometriosis imaging. They will also test it on larger, more diverse datasets from multiple centers to improve classification accuracy.
