|
|
|
| Development and validation of a CT radiomic model for distinguishing between T1-2 and T3-4 gastric cancers |
| WANG Zhixuan 1， WANG Xiaoxiao 1， LU Chao 1， LU Siyuan 1， DING Yi 1， ZHANG Jiulou 2， JIANG Pengcheng 3， SHAN Xiuhong 1 |
| 1. Department of Radiology， Affiliated People′s Hospital of Jiangsu University， Zhenjiang Jiangsu 212002； 2. Artificial Intelligence Imaging Laboratory, School of Imaging, Nanjing Medical University， Nanjing Jiangsu 210029； 3. Department of General Surgery， Affiliated People′s Hospital of Jiangsu University， Zhenjiang Jiangsu 212002， China |
|
|
|
| Guide |
|
Abstract Objective： To investigate the value of constructing a model based on CT radiomics for preoperative identification of T1-2 and T3-4 gastric cancers. Methods： A total of 465 gastric cancer patients with preoperative abdominal CT enhanced scanning and had clear T staging after resection were retrospectively recruited, and they were divided into two stages: T1-2 and T3-4. The patients were divided into training set and test set according to 7 ∶3 using stratified sampling method. The ROI was delineated and the radiomic features were extracted on the CT images of the venous phase. LASSO regression was adopted to screen out the features with the highest correlation with T staging, and logistic regression, support vector machine and decision tree were used to build the radiomics model. Radiomics signature were established based on radiomics features, clinical model was established based on clinical features, and a radiomics nomogram was constructed combining radiomics signature and clinical features. The receiver operating characteristic (ROC) curve was employed to evaluate the performance of the model in distinguishing T staging.Delong test compared the difference of area under ROC curve between optimal radiomic model and clinical model,as well as the difference between the nomogram and the more efficient model of the two.The calibration curve was utilized to evaluate the match between the model evaluation and the actual pathological results, and the decision curve was used to evaluate the net clinical benefit of models. Results： Among the radiomics models, the logistic regression model has the best predictive performance, with AUC values of 0.864 and 0.836 on the training set and test set, both higher than the clinical model. The radiomics nomogram, with the training AUC of 0.876 and test AUC 0.850, combined with radiomics signature and clinical features had better predictive performance than other models. Delong test showed that the predictive efficacy of logistic regression model was better than that of clinical model in the training set (P< 0.01). Although the AUC value of nomogram was higher than that of logistic regression model, the difference was not statistically significant.The calibration curve and decision curve reflected that the radiomics nomogram had good model adaptability and clinical utility. Conclusion： The nomogram constructed based on CT radiomics can provide important value for preoperative differential diagnosis of T1-2 and T3-4 gastric cancers.
［Key words］gastric cancer; CT; T staging; radiomics
|
|
|
| Fund: |
|
|
|
|
|
|
