CT-Based Radiomics to Differentiate Benign from Malignant Intraductal Papillary Mucinous Neoplasm
Objective This study aimed to explore the value of analysis based on CECT radiomics in assessing intraductal papillary mucinous neoplasm(IPMN).Methods A total of 51 patients with IPMN confirmed by pathology were included.They were divided into high-risk group and low-risk groups according to pathological grading.Preoperative CECT images and relevant clinical datas were collected for univariate and multivariate regression analysis.Textural features were extracted.Lasso-Logistic regression was used for feature selection,then Logistic regression prediction models were established.The generalization ability of the model was evaluated using ten-fold cross-validation,and the prediction ROC curve was drawn.Results Only one indicator of clinical and general imaging features,the enhanced mural nodule(OR=5.980,95％CI:1.678～21.311),is selected into the prediction model.Based on the images in arterial and venous phases,2 and 3 texture features were selected respectively.Prediction models were constructed separately.The AUC of the clinical radiological feature model(established by clinical and conventional imaging features),the CT arterial textural feature model(established by only arterial texture features),the CT venous textural feature model(established by only venous texture features),the arterial combined model(established by arterial texture features and clinical,conventional imaging features),the venous combined model(established by venous texture features and clinical,conventional imaging features)respectively was 0.667,0.801,0.830,0.847,0.859.Conclusion CECT-based radiomics is helpful for distinguishing malignant IPMN from benign IPMN,and the venous combined model has best performance.
万方数据
维普
