An emerging artificial intelligence tool called EndoFusion can detect signs of endometriosis in pelvic scans in just 18 milliseconds, according to a study published in the journal Artificial Intelligence in Medicine. Developed by researchers from Adelaide University’s Robinson Research Institute and the Australian Institute for Machine Learning, the framework combines data from magnetic resonance imaging (MRI) and transvaginal ultrasound scans to shorten a diagnostic timeline that currently averages seven years.
How EndoFusion Detects Endometriosis Through Single Scans
EndoFusion processes pelvic scans in less than a second by synthesizing data from multiple imaging modalities. According to Adelaide University researchers, MRI and ultrasound imaging each possess distinct strengths in identifying major indicators of advanced endometriosis, but patients typically only undergo one type of scan depending on local access and clinician preference. Jodie Avery, Associate Professor, Research Co Lead, Chronic Reproductive Conditions, Endometriosis Research Group, Robinson Research Institute, Adelaide University, stated that relying on a single, less optimal imaging tool can disadvantage patients, particularly when scan efficacy depends heavily on operator experience and cost constraints.
To overcome these hurdles, the AI tool pools data from four distinct datasets containing more than 9,000 female pelvic MRI scans and over 800 transvaginal ultrasound sliding scans. Lead author Dr. Yuan Zhang noted that testing showed the framework correctly distinguished between positive and negative cases of endometriosis 83% of the time, outperforming competing diagnostic models. The system aims to provide clinicians with a reliable, non-invasive method to determine the presence of the condition without requiring exploratory surgery.
Addressing the Challenges of Current Diagnostic Pathways
Endometriosis affects more than 190 million women worldwide. Historically, definitive diagnosis has required visual identification of lesions through invasive surgery—a procedure that carries surgical risks, high financial costs, and significant delays for patients.
Associate Professor Avery emphasized that establishing accurate, non-invasive early diagnostic tools remains critical for reducing the lengthy diagnostic timeline and cutting associated healthcare expenses. The research team partnered with multiple institutions for the project, including Flinders University, Benson Radiology, Omni Ultrasound and Gynaecological Care, the University of Surrey, McMaster University Medical Centre, and the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI).
Next Steps for Clinical Integration
The EndoFusion framework remains in the early stages of development. Researchers plan to expand the training dataset to incorporate additional diagnostic markers, which will further refine the tool’s classification accuracy. Dr. Zhang stated that future clinical deployment will allow physicians to evaluate multi-modal imaging data from a single patient encounter to determine the likelihood of endometriosis swiftly.
Beyond reproductive health, study authors suggest the underlying multi-modal AI approach could eventually offer diagnostic insights for other complex medical conditions that rely on combined imaging techniques. Potential future applications span other gynaecological disorders, prostate and breast cancer evaluations, and the detection of fetal abnormalities.