Background: Occult peritoneal metastasis (OPM) in advanced gastric cancer (AGC) patients remains a major diagnostic challenge. The aim of this study was to develop novel predictive models for identification of OPM in AGCs.
Method: A total of 810 patients with primary AGCs from two hospitals were retrospectively selected and divided into training (n = 393), internal validation (n = 215) and external validation cohorts (n = 202). CT based machine learning models were built and tested to predict the OPM status in AGCs., which are 1) Radiomic signatures: using venous CT imaging features, 2) Clinical models: integrating tumor location, differentiation and extent of serosal exposure, and 3) Radiomics models: combining of radiomic signature, tumor location and tumor differentiation.
Result: Total incidence of OPM was 8.27% (67/810). Clinical models yielded comparable classification accuracy with the corresponding radiomics models with similar AUCs (0.902-0.969 vs. 0.896-0.975) while the radiomic signatures showed relatively low AUCs of 0.863-0.976. In the case where the specificity is higher than 90%, the overall sensitivity of clinical model and radiomics model for OPM positive cases was 76.1% (51/67) and 82.1% (55/67). A nomogram based on the logistic clinical model was drawn to facilitate the usage and verification of the clinical model.
Conclusion: Both the novel CT based clinical nomogram and radiomics model provide promising method to yield high accuracy in identification of OPM in AGC patients.
Keywords: Computed tomography; Gastric cancer; Occult peritoneal metastasis; Radiomics.
Copyright © 2022 Elsevier Ltd, BASO ~ The Association for Cancer Surgery, and the European Society of Surgical Oncology. All rights reserved.