Objectives Ultrasound(US) is a cheap, non-invasive and well-recognized image modality for diagnosing and assessing ovarian cancer(OC). However, approximately 18% to 31% of adnexal lesions detected on US remain indeterminate. Machine learning(ML) is a promising tool for the implementation of complex multi-parametric algorithms. Despite the standardization of features capable of supporting the discrimination of ovarian masses into benign and malignant, there is the lack of accurate predictive modeling based on US examination for progression-free survival (PFS).

Methods In this retrospective observational study, we analyzed patients with epithelial ovarian cancer(EOC) who were followed in a tertiary center from 2018 to 2019. Demographic, clinical and laboratory characteristics were collected as well as information about post-surgery histopathology. Furthermore, we recorded data about US examinations according to International Ovarian Tumor Analysis(IOTA) classification. Proper feature selection was used to determine an attribute core set. Random Forest(RFF) algorithm was trained and validated with 10-fold cross-validation to predict 12-month PFS. The accuracy of the algorithm was than assessed scoring accuracy and Area Under Receiver Operating Characteristic(AUROC).

Results Our analysis included n.32OC patients with mean age of 54.1±14.9 years at diagnosis. Histotypes were n.19/32(59.4%) serous carcinoma, n.5/32(15.6%) mucinous, n.5/32(15.6%) endometriod and n.3/32(9.4%) clear cell. All patients underwent radical surgery. The attribute core set used to train machine learning algorithms is reported in figure 1. RFF showed an accuracy of 0.81, AUROC 0.91.

Conclusions We developed an accurate model to predict 12-month PFS in patients with OC based on a ML algorithm applied to gynecological ultrasound evaluation, requiring few easy-to-collect attributes.