Urine and Serum miRNA Signatures for the Non-Invasive Diagnosis of Adenomyosis: A Machine Learning-Based Pilot Study

Abstract

Background: Adenomyosis remains difficult to diagnose non-invasively due to clinical overlap with endometriosis and the limited specificity of imaging techniques. This pilot study evaluated whether serum- and urine-derived microRNA (miRNA) profiles, combined with machine-learning approaches, could support non-invasive diagnosis. Methods: Serum and urine samples were collected from 59 patients undergoing surgery for chronic pelvic pain at the Endometriosis Centre of RWTH Aachen University Hospital. Seven patients had isolated adenomyosis, 34 had histologically confirmed endometriosis, and 18 served as negative controls. miRNAs were profiled using next-generation sequencing. A structured feature-selection pipeline (variance filtering, univariate testing, mutual information, recursive feature elimination) was applied before training Logistic Regression, Random Forest, Support Vector Machine, and Decision Tree models using cross-validation. Model performance was evaluated using accuracy, precision, recall, F1 score, and ROC-AUC. Results: Distinct miRNA signatures were detected in both serum and urine, with urine-based models showing superior discriminatory performance. Logistic Regression and Support Vector Machine achieved excellent separation in urine datasets, although perfect AUC values must be interpreted cautiously due to the small number of adenomyosis cases. In serum, Random Forest achieved the highest AUC values (up to 0.98). Several miRNAs, including miR-183-3p, miR-320d-2, and miR-17, emerged as promising candidate biomarkers for differentiating adenomyosis from endometriosis and from negative controls. Conclusions: This pilot study demonstrates the feasibility of liquid-biopsy miRNA profiling combined with machine learning for non-invasive adenomyosis detection. Although results are preliminary and require validation in larger cohorts, urine miRNA profiles may represent a promising complementary tool to improve diagnostic accuracy and reduce diagnostic delay.