Development and validation of models to predict serosal invasion in advanced gastric cancer using the enhanced CT imaging-based radiomics features and clinical features / 中国医学物理学杂志

ABSTRACT

Objective To explore the predictive value of the enhanced CT imaging-based radiomics model and the clinical model for the serosal invasion in advanced gastric cancer.Methods The data were collected from 351 patients with advanced gastric cancer who underwent abdominal enhanced CT examination within 2 weeks before surgery,and the patients were randomly divided into a training group(n=247)and a validation group(n=104)in a ratio of 7:3.The 3190 radiomics features which were extracted from the arterial and venous phase CT images using A.K software were dimensionally reduced for constructing a radiomics model.The pathological features between serosal invasion positive and negative groups were compared,and the significant features were used to establish a clinical model.The model's performance was evaluated using receiver operating characteristic curve.Results In the training and validation groups,N staging and M staging were different in serosal invasion positive and negative groups(P<0.05).A total of 14 radiomic features were ultimately selected from the arterial and venous phase images.In the validation group,the diagnostic efficacy of the radiomic model for predicting serosal invasion in advanced gastric cancer was higher than that of the clinical model based on the combination of N staging and M staging(AUC:0.854 vs 0.793).Conclusion Both the radiomics model based on the enhanced CT imaging and the clinical model based on the combination of N staging and M staging can successfully predict serosal invasion in advanced gastric cancer,but the former performs better.