| Purpose:To develop and validate an effective model to predicte the resectability of pancreatic cancer.Materials and methods:Between January 2016 and June 2019,188 patients with pancreatic cancer confirmed by pathology and performed enhanced CT were enrolled from three grade A class 3 hospital and divided into primary and validation cohorts(131 in the primary cohort and 57 in the validation cohort).There are respectively 50 and 15 patients with unresectable pancreatic cancer in each cohort.Three models were built to predicte the resectability of pancreatic cancer,including a model based on radiomic signature alone,one based on clinic-radiological features alone and one that combined the two.The diagnostic performance of the three models was estimated and compared with the area under the receiver operating characteristic curve(AUC)in the validation cohort.Results:In the 188 patients,seven radiomic features based on enhanced CT images were selected after Boruta test to built the prediction model.In the primary cohort,the AUC for diagnosing the resectability of pancreatic cancer were 0.740,0.844 and 0.911 with the clinic-radiological model,the radiomics model and the combined model respectively.In the validation cohort,the AUC for diagnosis were 0.602,0.809 and 0.896 with the clinic-radiological model,the radiomics model and the combined model respectively.In the validation cohort,the diagnosis performance of the combined model was significantly improved over the model based on clinic-radiological features(AUC 0.896 vs 0.602,p value=0.001)and was comparable to the model based on the radiomic features(AUC 0.896 vs 0.809,p value=0.236).Conclusion:In predicting the resectability of pancreatic cancer,the combined model is superior to the model based on clinic-radiological features alone and is comparable to the model based on the radiomics features alone.The combined model can provide added information for the therapeutic choice of pancreatic cancer. |
PDF Full Text Request