Introduction/Background Women with endometriosis (EMS) present an increased risk of developing EMS-related ovarian carcinoma (EROC). To date, the definitive diagnosis relies on the post-surgical assessment by a pathologist, thus limiting individualized medical approaches due to the lack of screening methods for the timely detection of EROC development. To fill in this unmet clinical need, the purpose of this study was to explore the capacity of radiomic-based machine learning (ML) models to discriminate EROC from EMS lesions and normal ovaries using pre-operative contrast-enhanced CT (CE-CT) images.

Methodology Volumes of interest from both ovaries were semi-automatically contoured by an experienced radiologist. Feature pre-processing, harmonization, and extraction were done using the PyRadiomics library. Three ML models were implemented in R for the binary classification task, employing a 10-fold cross-validation approach with a 70:30 ratio between training and test sets. The ML models were based on different functions for feature reduction (i.e., Boruta, Recursive Feature Elimination, and Least Absolute Shrinkage and Selection Operator) and integrated with generalized linear models. The predictive ability of each classifier was assessed using area under the curve (AUC), sensitivity and specificity.

Results In the 95 prospectively enrolled women (median[range] age: 58[31–79] years) with definitive diagnosis, 45 EROC-related ovaries and 65 non-tumoral controls were delineated, while 80 volumes were excluded being borderlines or with other cancer types. The ML models, each based on 1409 extracted radiomic features (RFs), showed good performances with AUC, sensitivity, and specificity ranging from 0.72 to 0.88, from 0.54 to 0.92, and from 0.84 to 0.89, respectively.

Conclusion ML models based on CE-CT-derived RFs have great potential in differentiating EROC-related patients against controls, thus representing a valuable and objective information to complement the clinical decision in the attempt to implement fertility sparing approaches.

Disclosures The study is funded by the Italian Ministry of Health (ENDO-2021–12371926).