A CT-Based Radiomics Model for Evaluating Peritoneal Cancer Index in Peritoneal Metastasis Cases: A Preliminary Study

The present work aimed to develop and validate a radiomics model for evaluating peritoneal cancer index (PCI) in peritoneal metastasis (PM) cases based on preoperative CT scans.

Pathologically confirmed pancreatic, colon, rectal, and gastric cancer cases with PM administered exploratory laparotomy in 2 different cohorts were retrospectively analyzed. Surgical PCIs (sPCIs) were confirmed by the surgery team, and CT-PCI scores were assessed by radiologists. Totally 63 and 27 cases in cohort 1 were assigned to the training and test groups, respectively. Then, 73 cases in cohort 2 were enrolled as an external validation set. Radiomics features were derived from the portal venous phase of preoperative abdominopelvic CT scans. Nineteen optimal features related to sPCI were finally selected. Support vector machine (SVM) was adopted for radiomics model generation. The associations of CT-PCI, radiomics PCI and sPCI were analyzed. The performances in distinguishing between low-sPCI (≤ 20) and high-sPCI (> 20) cases were also assessed by receiver operating characteristic (ROC) curve analysis and decision curve analysis (DCA).

Both CT-PCI and radiomics PCI scores had positive associations with sPCI. The radiomics approach had higher agreement for detecting sPCI than CT-PCI. In addition, the radiomics model had enhanced diagnostic performance than CT-PCI (AUCs were 0.894, 0.822 and 0.810 in training, test and validation sets, respectively, vs 0.749, 0.678 and 0.693, respectively). The net reclassification index was 0.266. The usefulness of the proposed model was confirmed by DCA in an external validation set.

The present pilot study showed that the radiomics model based on preoperative abdominopelvic CT has increased agreement and diagnostic performance in detecting sPCI than CT-PCI in patients with PM, which could be used to optimize individualized treatment strategies.