Machine learning and deep learning for diagnosis of Polyendocrine Metabolic Ovarian Syndrome: systematic review and...

Polyendocrine Metabolic Ovarian Syndrome (PMOS) is a prevalent endocrine disorder with a challenging, heterogeneous diagnosis. Machine learning (ML) and deep learning (DL) models show promise for automated diagnosis, but a quantitative synthesis of their accuracy is lacking.

To evaluate the diagnostic accuracy of ML/DL models for PMOS and identify factors influencing performance.

We searched MEDLINE, Web of Science, Embase, and Cochrane Library from inception to January 26, 2026. Studies developing or validating ML/DL models for PMOS diagnosis were included. Risk of bias was assessed using QUADAS-2. A random-effects model with the Hartung-Knapp-Sidik-Jonkman method was used to pool estimates. Hierarchical summary receiver operating characteristic curves were constructed. Heterogeneity was quantified (I ²), and meta-regression and subgroup analyses explored sources of heterogeneity.

Fifty-six studies (60 datasets) were included. Pooled sensitivity was 0.92 (95% CI: 0.89-0.94; I²  = 93.6%) and specificity 0.94 (95% CI: 0.91-0.96; I ²  = 93.4%), with an HSROC area under the curve of 0.99 (95% CI: 0.97-0.99). Prediction intervals were wide (sensitivity: 50%-99%; specificity: 53%-100%). Egger's test suggested small-study effects (P < 0.05). Meta-regression identified data source (database vs. single-center, coefficient=1.91, P = 0.002) and modeling variables (ultrasound image vs. clinical parameters, coefficient=2.43, P = 0.007) as independent predictors of higher accuracy; model type was not significant after adjustment. Ultrasound image-based models achieved the highest accuracy (sensitivity 0.99, specificity 0.98). Notably, 32 studies did not report diagnostic criteria, and only 4 performed external validation.

ML/DL models, particularly those using ultrasound imaging, demonstrate promising but conditional diagnostic accuracy for PMOS. However, poor reporting of diagnostic criteria, lack of external validation, and substantial heterogeneity limit current evidence. Future research must prioritize rigorous validation and adherence to reporting standards. While not yet ready for independent clinical use, these models hold promise as assistive tools to standardize ovarian assessment.

https://www.crd.york.ac.uk/PROSPERO/view/CRD420251137107, identifier CRD420251137107.