Objective To investigate the predictive value of enhanced computed tomography (CT) texture analysis for Claudin 18.2 expression in gastric cancer.

Methods The retrospective cohort study was conducted. The clinicopathological data of 196 patients with gastric cancer who were admitted to Jiangsu Province Hospital and Nanjing BenQ Medical Center of Nanjing Medical University from January 2023 to October 2025 were collected. There were 112 males and 84 females, aged (59±12) years. Patients were randomly divided into a training set (137 cases) and a validation set (59 cases) at a ratio of 7∶3 using the random number table method. The training set was used to construct the predictive model, and the validation set was used to validate the predictive model. All patients underwent preoperative abdominal enhanced CT examination, and received radical gastrectomy or gastroscopy to obtain specimens. Observation indicators: (1) Claudin 18.2 expression and comparison of clinicopathological characteristics; (2) results of texture feature screening; (3) construction and efficacy evaluation of the predictive model. Comparison of measurement data with normal distribution between groups was conducted using the independent sample t test. Comparison of measurement data with skewed distribution between groups was conducted using the Mann‑Whitney U 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 rank sum test. Intraclass correlation coefficient (ICC) was used for consistency test, and ICC >0.75 indicated good consistency. Receiver operating characteristic (ROC) curves were plotted, and the area under the curve (AUC), calibration curve, and decision curve were used to evaluate the predictive efficacy of each model. DeLong test was used to compare the AUC.

Results (1) Claudin 18.2 expression and comparison of clinicopathological characteristics: among 196 gastric cancer patients, 63 cases were Claudin 18.2 positive and 133 cases were negative. There were 43 and 20 positive cases in the training and validation sets respectively, with no significant difference in the positive proportion between the two groups (χ²=0.12, P>0.05). In the training set, cases with Lauren classification as intestinal type, diffuse type, and mixed type were 10, 21, and 12 for Claudin 18.2 expression positive gastric cancer patients, versus 36, 25, and 33 for Claudin 18.2 expression negative gastric cancer patients. Cases with well‑differentiated, moderately differentiated, and poorly differentiated tumor were 5, 8, and 30 for Claudin 18.2 positive gastric cancer patients, versus 15, 41, and 38 for Claudin 18.2 negative gastric cancer patients. There were significant differences in the above indicators between them (χ²=6.80, Z=-2.80, P<0.05). (2) Results of texture feature screening: a total of 156 texture features were extracted from the arterial and venous phases of enhanced CT examination. Among them, 132 texture features with inter‑observer and intra‑observer ICC >0.75 were included in least absolute shrinkage and selection operator regression analysis for further screening. Nine key texture features associated with Claudin 18.2 expression were identified, including 5 from the arterial phase and 4 from the venous phase, with non‑zero feature coefficients and good stability. (3) Construction and efficacy evaluation of the predictive model: based on clinical characteristics and CT imaging features, a clinical prediction model, a CT texture feature prediction model, and a combined prediction model were constructed. In the training set, the AUCs of the clinical prediction model, CT texture feature prediction model, and combined prediction model were 0.69 95% confidence interval (CI) as 0.60-0.76, 0.89 (95%CI as 0.83-0.94), and 0.90 (95%CI as 0.84-0.94), respectively. In the validation set, the AUCs of the prediction three models were 0.71 (95%CI as 0.61-0.87), 0.88 (95%CI as 0.77-0.96), and 0.89 (95%CI as 0.78-0.96), respectively. The results of DeLong test showed that in the training set, the AUCs of the combined prediction model and the CT texture feature prediction model were both higher than that of the clinical prediction model (Z=4.56, 3.82, P<0.05). The results of calibration curve analysis showed a good calibration performance in the combined prediction model. The results of decision curve analysis showed clinical application value in the combined prediction model.

Conclusion Eenhanced CT‑based texture analysis can effectively predict Claudin 18.2 expression status in gastric cancer, and the combined prediction model incorporating clinicopathological characteristics has better predictive efficacy than single models.