Preliminary Study On Preoperative T Staging Of Rectal Cancer Based On Radiomics Of High Resolution T2-weighted Imaging

Objective:To explore the application value of radiomics features extracted from high-resolution T2-weighted imaging(T2WI)for preoperative T staging of rectal cancer.Methods:191 patients with pathologically confirmed rectal cancer in Shanxi Provincial People's Hospital from March 2016 to March 2018,without treating neoadjuvant radiotherapy and chemotherapy.The whole tumor volume was delineated on high-resolution T2 WI by radiologists,the imaging features were extracted by Artificial Intelligence Kit Software.The Least Absolute Shrinkage and Selection Operator(LASSO)was used to reduce redundancy features and the radiomics model was build.The modeling process is mainly divided into four steps.The first was data preprocessing.The second was the features reduction dimension.A total of 396 features were extracted,and the selected features were 14.The third was the model establishment and the Rad-score's getting.The Random Forest(RF)algorithm was first used to establish a radiomics model,and then the Rad-score generated by radiomics was combined with clinical data to model;the fourth was model evaluation(data validation).The patients were divided into two groups:early stage(T1~T2)and local advanced stage(T3~T4).The pathological results were used as the true value to train and test the model.The receiver operating characteristic curve(ROC)and decision curve analysis(DCA)were plotted to assess the predictability of the model.The accuracy,sensitivity,and specificity of conventional MRI assessments werecompared to those obtained from predictive models.Results:Among the 191 patients,68 cases were in the early stage,and the local advanced stage was 123 cases.The diagnosis of T stage of the predictive model in training and testing were: Area Under Curve(AUC): 0.927,0.885,accuracy: 0.851,0.842,specificity: 0.917,0.7,sensitivity: 0.814,0.919,F1: 0.861,0.829,Brier:0.149,0.159.The accuracy,sensitivity,and specificity of the T-stage diagnosis of conventional MRI by professional doctors were 0.815,0.865,and 0.75.The confusion matrix obtained by the model showed that the positive predictive value and the negative predictive value in the training group were 0.814 and 0.917,respectively;the positive predictive value and the negative predictive value in the testing group were 0.919 and 0.7.The DCA of the model in the testing group showed that while the probability of predicting local advanced rectal cancer ranging from 40%to 95%,the method to predict the local advanced rectal cancer was superior than either the scheme in which all patients were assumed to local advanced rectal cancer or the scheme in which all patients were early rectal cancer.The model has a significant clinical benefit in assessing preoperative T stage of rectal cancer.Conclusion:This radiomics model constructed by high-resolution T2 WI to determine the T stage of rectal cancer is feasible,and can provide a better model than the MRI evaluation.