Artificial intelligence for the automated detection of polycystic ovary syndrome on ultrasonography

Polycystic ovary syndrome (PCOS) affects 1 in 8 women of reproductive age imparting serious reproductive, cardiometabolic and psychosocial health complications for patients across the lifespan. Despite a high prevalence, under- and misdiagnosis of PCOS is common and contributes to significant delays in the delivery of evidence-based care. Lack of standardization in the clinical evaluation of PCOS features contributes to these delays in diagnosis. We have shown that ultrasonographic aspects of ovarian morphology have high diagnostic accuracy for PCOS and recently provided the new international standards for defining polycystic ovaries on ultrasonography. However, conventional assessments of ovarian morphology have moderate to poor inter-rater reliability which impacts their performance in clinical practice. Artificial intelligence provides an exciting opportunity to harmonize the ultrasonographic evaluation of PCOS. We propose that the use of deep- learning approaches for image classification, segmentation, and integration with clinical and biochemical markers can provide a framework to enable the timely detection and treatment of this highly prevalent condition. To that end, this project will develop a model for the automated characterization of polycystic ovarian morphology on ultrasonography (Aim 1), that can be leveraged alongside clinical and biochemical parameters to detect PCOS status (Aim 2). Our approach involves the use of a highly unique archive of ultrasonographic images and volumes of the ovaries garnered from well-characterized women across the reproductive spectrum, as well as external validation of the model for PCOS status using 2-dimensional and 3-dimensional views of ovarian morphology. The hypotheses to be tested are that a trained and externally validated deep learning model can detect follicle excess, ovarian enlargement and/or stromal aberrations that effectively define polycystic ovarian morphology, and that integration of one or more clinical or biochemical parameters yields a high performing model for the classification of PCOS status. By providing a model for the automated detection of PCOS, we expect to make important strides toward resolving subjectivity in the clinical evaluation of PCOS. Ultimately, standardization in the diagnostic assessment is needed to obviate delays in detection and facilitate access to treatment and prevention strategies that improve the overall health and well-being of patients living with PCOS. Project Number: 1R21HD116139-01A1 | Fiscal Year: 2025 | NIH Institute/Center: Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD) | Principal Investigator: Marla Lujan (+2 co-PIs) | Institution: CORNELL UNIVERSITY, ITHACA, NY | Award Amount: $245,157 | Activity Code: R21 | Study Section: Emerging Imaging Technologies and Applications Study Section[EITA] View on NIH RePORTER: https://reporter.nih.gov/project-details/1R21HD11613901A1