Objective To investigate the clinical value of a combined prediction model based on dual-energy computed tomography (CT) features in vascular invasion of gastric cancer.

Methods The retrospective cohort study was conducted. The clinicopathological data of 368 gastric cancer patients who were admitted to Henan Provincial People's Hospital from December 2020 to December 2024 were collected. There were 258 males and 110 females, aged 63(range, 50-73) years. Among the 368 patients, 199 cases had vascular invasion, and 169 cases had no vascular invasion. Observation indicators: (1) consistency analysis of CT examination; (2) comparison of clinical characteristics between gastric cancer patients with and without vascular invasion; (3) multivariate analysis of vascular invasion in gastric cancer patients; (4) construction of prediction model and assessment of diagnostic efficacy. Comparison of measurement data with normal distribution between groups was conducted using the independent sample t test. Comparison of count data between groups was conducted using the chi-square test. Comparison of ordinal data between groups was conducted using the Mann-Whitney U test. The consistency in CT findings and CT numerical variables between two radiologists was evaluated using the Kappa test and intraclass correlation coefficient (ICC). Univariate analysis was performed using appropriate statistical methods based on data type, and multivariate analysis was conducted using Logistic regression model. The efficacy of the prediction model was evaluated using receiver operating characteristic (ROC) curve, and the area under the curve (AUC), sensitivity, specificity, and Youden index were calculated. Comparison between prediction models was conducted using the Delong test.

Results (1) Consistency analysis of CT examination: the Kappa value of imaging T stage measured by the two radiologists was 0.9095% confidence interval (CI) as 0.83-0.94. The ICC for tumor thickness, tumor volume, portal vein phase 70 keV CT (PvP-70keV CT) value, iodine concentration in the portal vein phase lesion (PvP-IC), normalized iodine concentration in the portal vein phase (PvP-NIC) measured by the two physicians were all >0.80. (2) Comparison of clinical characteristics between gastric cancer patients with and without vascular invasion: for patients with vascular invasion, the tumor thickness, tumor volume, cases with tumor differentiation degree as poorly-, moderate to well-differentiated, cases of stage Ⅰ-Ⅱ or stage Ⅲ of imaging TNM staging, portal venous phase dual-energy CT parameters (PvP 70 keV CT value, PvP-IC, PvP-NIC), cases with high or low human epidermal growth factor receptor 2 (HER2) expression, cases with high or low Ki-67 expression were (1.6±0.4) cm, (34±7) cm³, 99, 100, 91, 108, (82±11) HU, (2.6±0.4) mg/mL, 0.43±0.09, 51, 148, 125, 74, respectively. For patients without vascular invasion, the above indicators were (1.4±0.3) cm, (28±6) cm³, 59, 110, 99, 70, (74±9) HU, (2.3±0.3) mg/mL, 0.35±0.06, 28, 141, 81, 88, respectively. There were significant differences in the above indicators between them (t=4.68, 9.71, χ²=8.21, 6.04, t=8.00, 8.02, 9.85, χ²=4.45, 8.22, P<0.05). (3) Multivariate analysis of vascular invasion in gastric cancer patients: results of multivariate analysis showed that imaging TNM staging, PvP-70 keV CT value, PvP-IC, PvP-NIC, HER2 expression, and Ki-67 expression were independent influencing factors for vascular invasion in gastric cancer patients (odds ratio=2.39, 1.19, 1.20, 1.20, 2.13, 1.34, 95%CI as 1.04-5.49, 1.04-1.36, 1.05-1.36, 1.06-1.35, 1.20-3.78, 1.07-1.68, P<0.05). (4) Construction of predictive model and diagnostic efficacy evaluation: a predictive model was constructed based on results of multivariate analysis. The ROC curve showed that the AUC of the imaging parameters combined with clinical indicators predictive model was 0.92(95%CI as 0.87-0.98), with sensitivity of 0.94, specificity of 0.78, and Youden index of 0.72. The above indicators were 0.92(95%CI as 0.86-0.98), 0.92, 0.78, 0.70 for imaging parameters combined predictive model, 0.78(95%CI as 0.67-0.88), 0.64, 0.81, 0.45 for PvP-70 keV CT value predictive model, 0.79(95% CI as 0.69-0.90), 0.75, 0.76, 0.51 for PvP-IC predictive model, 0.81(95%CI as 0.71-0.92),0.76,0.80, 0.56 for PvP-NIC predictive model, 0.85(95%CI as 0.75-0.94), 0.76, 0.92, 0.68 for HER2 expression predictive model, 0.80(95%CI as 0.69-0.90), 0.75, 0.78, 0.53 for Ki-67 expression predictive model, respectively. The diagnostic efficacy of the imaging parameters combined with clinical indicators predictive model was significantly higher than that of the predictive models based on PvP-70keV CT value, portal venous phase PvP-IC, portal venous phase PvP-NIC, human epidermal growth factor receptor 2 expression, and Ki-67 expression (Z=4.32, 3.98, 3.76, 2.89, 3.46, P<0.05).

Conclusions Imaging TNM stage, PvP-70keV CTvalue, PvP-IC, PvP-NIC, human epidermal growth factor receptor 2, Ki-67 expression are independent influencing factors for vascular invasion in gastric cancer patients. The predictive model constructed based on the above factors demonstrates high clinical value in predicting vascular invasion in gastric cancer.