The Risk Stratifications And Prognositic Analysis In Patients With Thymic Epithelial Tumors Based On CT Imaging And Random Forest Algorithm

Part One:Risk stratifications in patients with thymic epithelial tumors based on CT imaging and random forest algorithmObjective:To determine the predictive features of thymic epithelial tumors using random forest algorithm.Methods and Materials:A total of 169 patients with pathologically confirmed thymic epithelial tumors were enrolled in this study.A total of 3 clinical features and 21 CT features were reviewed.The association between CT features and risk stratifications were analyzed by random forest algorithm,and the predictive efficiency was evaluated by ROC analysis.Results:Among the 169 patients,there were 32 low-risk thymomas,45 high-risk thymomas and 92 thymic carcinomas in this study.Random forest suggested that tumor shape,septum,mediastinal invasion,lymphadenopathy and pericardial effusion were predictive features in thymic epithelial tumors.Tumor shape,lymphadenopathy and pericardial effusion were the key features for predicting high-risk thymomas and thymic carcinomas,with area under curve was 0.957 in ROC analysis.Patient age,tumor size,necrosis or cystic change and lung metastasis were the key features for predicting low-risk thymomas and high-risk thymomas,with area under curve was 0.980 in ROC analysis.Conclusions:The random forest model in the present study has high efficiency in predictive diagnosis of thymic epithelial tumors.Thymic tumors with irregular shape,the presence of lymphadenopathy and pericardial effusion are highly indicative of thymic carcinomas.The presence of necrosis and lung metastasis were more frequently seen high-risk thymomas than in low-risk thymomas,and the tumor size was bigger and patient age was younger in high-risk thymomas than low-risk thymomas,which may help in tumor differentiation.Part Two:Prognostic CT features in patients with thymic epithelial tumorsObjective:To determine the prognostic CT features in patients with untreated thymic epithelial tumors.Methods and Materials:A total of 195 patients with pathologically confirmed TETs were enrolled in this study.The clinical and survival data were collected.The following-up periods since first CT examiniations were at least for 3 years except for the missing data.Sixteen CT features were respectively reviewed.The association between CT features and patients’ survival status was analyzed by Cox regression.Results:There were 110 thymic carcinomas,53 high-risk thymomas and 32 low-risk thymomas in this study.The following-up periods were between 22.5 to 144.3 months.Survival analysis shows the three year survival rate for high-riak thymomas and thymic carcinomas were 0.92 and 0.72,respectively.In the group of thymic carcinomas,33 patients(30.0%)died of thymic tumors or related diseases.Cox regression showed that lung invasion,great vessel invasion,lung metastases and distant organ metastasis were the independent predictors for patient survival(p<0.01).In the group of high-risk thymomas,8 patients(15.1%)died of thymic tumors or related diseases.Cox regression showed that lung invasion and pericardium mass were the independent predictors for patient survival(p<0.01).In the group of low-risk thymomas,2 patients(6.2%)died of non-thymic tumor related diseases,and no features were found related with patient survival.Conclusions:The present study revealed that CT could serve as an important tool to predict the survival outcomes in patients with TETs.