CT-based radiomics model to predict platinum sensitivity in epithelial ovarian carcinoma: a multicentre study

Abstract

<h3>Objective</h3><p>Platinum resistance carries poor prognosis in epithelial ovarian carcinoma (EOC). This study aimed to assess the value of radiomics model based on contrast-enhanced CT (ceCT) in predicting response to platinum-based chemotherapy in EOC.</p><h3>Materials and methods</h3><p>Patients with histologically confirmed EOC and pre-treatment ceCT were retrospectively recruited from 5 centres. All patients underwent standard platinum-based chemotherapy and optimal cytoreduction. Platinum sensitivity was determined by whether it recurred within six months after platinum-based chemotherapy. The whole tumour volume was manually segmented on the baseline ceCT. Radiomics features were extracted using the open-source package PyRadiomics (version 3.0.1). Patients from centres A-C were randomly divided into training and internal validation sets in 4:1 ratio. Patients from the centres D and E were assigned as independent external validation sets. Spearman’s rank correlation followed by 5-fold stratified cross validation (SCV) elastic net repeated for 100 times, and Mann-Whitney U test were deployed for feature reduction and selection. Adaptive synthetic sampling was applied to minimize class biases. Extra Trees classifier across 10-fold SCV was used for model building. The area under curve (AUC), calibration curve assessment, and decision curve analysis (DCA) were deployed to evaluate model performance and translational clinical utility.</p><h3>Results</h3><p>Seven hundred and three EOC patients (51.6 ± 9.3 years) were recruited. The training data (<em>n</em> = 608) yielded the following classification metrics: AUC (0.917), sensitivity (83.9%), specificity (94.4%), and accuracy (91.7%) in the internal validation set. The external validation set using centre D (<em>n</em> = 44) had AUC (0.877), sensitivity (76.5%), specificity (92.6%), and accuracy (86.4%); while centre E (<em>n</em> = 51) had AUC (0.845), sensitivity (73.3%), specificity (86.1%), and accuracy (82.4%) in predicting platinum sensitivity. DCA illustrated net clinical benefit in internal validation set and both external validation sets.</p><h3>Conclusions</h3><p>The proposed CT-based radiomics model could be useful in predicting platinum sensitivity in EOC with potential in guiding personalized treatment in EOC.</p>