To construct a prediction model based on CT radiomics features to predict the recurrence risk of gastric cancer after radical operation, and to explore the clinical value of its preoperative dynamic evaluation.

A total of 215 patients with gastric adenocarcinoma who underwent radical gastrectomy from January 2020 to December 2023 were retrospectively selected as the research objects. All patients completed multi-phase (plain scan, arterial phase and venous phase) enhanced CT examination before operation. Siemens 256-slice CT was used for standardized scanning, and the three-dimensional volume of venous tumor was manually delineated by ITK-SNAP software (the consistency between observers was Kappa>0.85), and 1152 imaging features were extracted based on PyRadiomics. The core features were screened by intra-observer and inter-observer consistency test (ICC≥0.75), single factor analysis and LASSO regression (λ=0.032, 10 times cross-validation) and a Logistic regression model was constructed. The data set was randomly divided into training set (150 cases) and independent verification set (65 cases) according to the ratio of 7∶3, and the prediction efficiency of the model was evaluated.

Among 215 patients with gastric cancer, 67 patients had postoperative recurrence, and the remaining 148 patients did not have postoperative recurrence. The results of univariate analysis showed that there were significant differences in vascular infiltration, nerve infiltration, tumor differentiation, surgical margin status and receiving adjuvant chemotherapy between patients with recurrence and those without recurrence (P<0.05). After a series of screening, the model finally retained eight core image omics features, including morphology, first-order statistics, texture analysis and wavelet transform features. Receiver operating characteristic curve analysis showed that the area under the curve (AUC) of the CT image omics model in predicting the training set was 0.865, and the sensitivity and specificity were 69.4% and 88.1% respectively. The AUC in predicting validation set was 0.875, and the sensitivity and specificity were 94.4% and 63.8% respectively.

The prediction model based on the characteristics of CT imaging shows good diagnostic efficiency in the risk prediction of recurrence after radical gastrectomy, and realizes the preoperative dynamic risk assessment of gastric cancer patients. This model can provide objective decision support for intensive follow-up and accurate adjuvant treatment of high-risk patients.