From Operating Room to Office: Translational Pathway for a Machine Learning–Driven Uterine Lavage Liquid Biopsy for Endometriosis and Adenomyosis

INTRODUCTION: Endometriosis and adenomyosis are leading causes of chronic pelvic pain and infertility, affecting millions of reproductive-aged women worldwide. Despite their prevalence, diagnostic delays average 8–10 years, driven by reliance on invasive laparoscopy and histologic confirmation. Imaging modalities lack sufficient sensitivity and specificity, especially for early or superficial disease. These limitations prolong patient suffering, contribute to unnecessary surgeries, and impose a $22–119 billion annual economic burden. An accurate, minimally invasive diagnostic platform that can be integrated into routine office care remains an urgent unmet need. OBJECTIVE: To develop and evaluate the diagnostic performance of a uterine lavage–based liquid biopsy platform, integrating extracellular vesicle (EV) proteomics and proprietary machine learning (ML) classification, for detection of endometriosis and adenomyosis in women presenting with pelvic pain. METHODS: Uterine lavage samples were collected during clinical procedures under IRB-approved protocols from multiple and independent health systems. EVs were isolated, and proteins purified and analyzed via mass spectrometry–based proteomics. A novel entropy-based ML algorithm was used to classify the resultant molecular signatures, minimizing batch effects and enabling the generation of compact diagnostic panels. Samples and proteomic profiles from 668 women were included in total, with diagnoses representing endometriosis, adenomyosis, leiomyomas, polyps, and controls. RESULTS: Using this platform, a series of iterations were performed that allowed us to select and validate a final <40-protein diagnostic classifier. For endometriosis/adenomyosis, performance included AUC 0.88–0.95, sensitivity 82–91%, specificity 84–88%, PPV50 90–97%, and NPV50 98–99%. Comparable results were observed for leiomyomas and polyps, with discrimination exceeding, in general, that of ultrasound and MRI. CONCLUSIONS: This work establishes the feasibility of a uterine lavage EV proteomic test for accurate diagnosis of endometriosis and adenomyosis. Results exceed imaging benchmarks and uniquely differentiate among common mimics of chronic pelvic pain. Current studies are based on lavage collected during clinical procedures; translation into routine practice is envisioned through an FDA-cleared, office-based collection device, enabling rapid, nearly painless sample acquisition during standard gynecologic visits. Together, this approach has the potential to reduce diagnostic timelines from years to days, decrease unnecessary operating room hysteroscopies, and alleviate the patient and health care burden. Future directions: Ongoing prospective multicenter trials will refine panel composition, validate clinical performance, and accelerate integration of this diagnostic into routine gynecologic care.