The Value Of CT-based Radiomics In Differentiating Gastric Neuroendocrine Carcinoma From Gastric Adenocarcinoma

Objective:To investigate the diagnostic value of a CT-based radiomics model in differentiating gastric adenocarcinoma(ADC)and gastric neuroendocrine carcinoma(NEC).Materials and methods:Forty-five patients with pathologically confirmed gastric neuroendocrine carcinoma and 99 patients with gastric adenocarcinoma from our hospital were retrospectively collected as the training set.Twenty-three patients with pathologically confirmed gastric neuroendocrine carcinoma and 40 patients with gastric adenocarcinoma from external hospitals were collected as the external validation set.Clinical information such as personal history and tumor markers were collected from the patients,and CT features were evaluated.Clinical and CT characteristics were screened for independent risk factors by univariate and multifactorial analyses,and clinical models were developed using binary logistic regression.Radiomics features were extracted and downscaled from the venous phase images of CT-enhanced scans.and the screened features were used to build Radiomics models.The diagnostic efficacy of the model was examined by plotting the receiver operating characteristic(ROC).Calculate the Rad-score of the radiomics model and construct a combined clinicalradiomics model by combining independent risk factors in the clinical features.Calibration curves were used to assess the goodness of fit of the multiple models.The differences in AUC values between the models were compared using the Delong test.Decision curve analysis(DCA)was used to assess the clinical application value of each model.Results:Univariate and multifactorial analysis showed that the tumor site,necrosis,and lymph node metastasis were independent risk factors for differentiating gastric adenocarcinoma from gastric NEC.A clinical prediction model was developed using binary logistic regression for the above 3 factors,and an AUC of 0.768(95%CI:0.6890.846)was obtained for the training set and 0.811(95%CI:0.7-0.921)for the validation set.The radiomics model contaioning nine features built by logistic regression had a training set AUC of 0.796(95%CI:0.716-0.876)and a validation set AUC of 0.813(95%CI:0.696-0.93).The combined model using clinical characteristics and Rad-score had the best diagnostic efficacy with an AUC of 0.874(95%CI:0.81-0.938)in the training set and 0.86(95%CI:0.756-0.963)in the validation set,which was statistically different from the radiomics histology model by Delong’s test.Conclusion:Both the clinical features model established by logistic regression analysis and the radiomics model based on enhanced CT had certain diagnostic abilities in the differentiation of gastric adenocarcinoma from gastric neuroendocrine carcinoma.The diagnostic efficacy could be further improved by using the combined model established by clinical and radiomics features.