The Application Of Radiomics Based On High-resolution MRI In The Diagnosis And Treatment Of Rectal Cancer

Part ? The Value of Radiomics Based on High-resolution Magnetic Resonance T2 WI in the Differential Diagnosis of Rectal LesionsPurpose: To explore the clinical application of radiomics based on high resolution T2 WI in the preoperative differentiating benign from malignant lesions of rectum.Methods: One hundred and forty-one pathologically confirmed patients with rectal lesions were enrolled and underwent 3-Tesla magnetic resonance high-resolution T2 WI of rectum from June 2016 to June 2018.The lesions were segmented on T2 WI to draw the outline of VOI.The radiomics features were extracted from the VOI,then the features were selected by least absolute shrinkage and selection operator(LASSO)method to best differentiate benign and malignant lesions.The samples are randomly divided into training set(80%)and test set(20%)for machine learning.Four classifier models of Logistic Regression(LR),Random Forest(RF),Decision Tree(DT)and K-nearest neighbor(KNN)were used to obtain the receiver operating characteristic(ROC)curves and calculate the area under the curve(AUC),sensitivity and specificity.Compare the difference of ROC curves with De Long test.Results: A total of 141 patients including 89 males and 52 females,age range from 26?78 years old,average age 53.2±11.3 years old.There were 24 benign lesions and 123 malignant lesions confirmed pathologically.1409 features were extracted from high-resolution T2 WI,and 8 features related to the identification of benign and malignant were obtained.The AUC of four classifier test group of LR,RF,DT and KNN were 0.802,0.779,0.590,0.733 respectively.For the LR classifier,the AUC was 0.802(95%CI: 0.633-0.887),sensitivity 76.4%,specificity 75.3%.LR classifier model has better diagnostic performance than the others(P < 0.05).Conclusion: The radiomics based on high-resolution T2 WI could provide important reference for differentiation between benign and malignant lesions of rectum,may help to assist clinical decision-making in treatment.Part ? The Value of Radiomics Based on High-resolution Magnetic Resonance T2 WI in the Preoperative Diagnosis of Rectal CancerPurpose: To explore the clinical application of radiomics based on high-resolution magnetic resonance T2 WI in the preoperative diagnosis of rectal cancer.Methods: One hundred and sixty-five pathologically confirmed patients with rectal cancer were enrolled and underwent 3-Tesla high-resolution magnetic resonance T2 WI of rectum from June 2016 to June 2019.The lesions were segmented on high resolution T2 WI to draw the outline of VOI.The radiomics features were extracted from the VOI,then the features were selected by least absolute shrinkage and selection operator(LASSO)method to best predict the pathological features.The samples are randomly divided into training set(80%)and test set(20%)for machine learning.Four classifier models of Logistic Regression(LR),Random Forest(RF),Decision Tree(DT)and K-nearest neighbor(KNN)were used to obtain the receiver operating characteristic(ROC)curves and calculate the area under the curve(AUC),sensitivity and specificity.Compare the difference of ROC curves with DeLong test.Results: A total of 165 patients including 109 males and 56 females,age range from 32~78 years old,average age 57.5±8.8 years old.1409 features were extracted from high-resolution T2 WI,and 15 features related to the T stage,10 features related to the N stage and 9 features related to the pathologic differentiation.The AUC of four classifier models of LR,RF,DT and KNN have good diagnostic performance for T stage,N stage and pathologic differentiation.For the RF classifier,the AUC of test group were 0.813(95%CI: 0.667-0.933,sensitivity 73.3% and specificity 72.2%),0.741(95%CI: 0.593-0.874,sensitivity 67.1% and specificity 75.0%),0.788(95%CI: 0.534-0.982,sensitivity 40.0% and specificity 96.4%),respectively for these three pathological features.RF classifier model has better diagnostic performance than the other three models(P < 0.05).Conclusion: The radiomics based on high-resolution MR T2 WI could differentiate pathologic differentiation,T stage and N stage,indicating that radiomics may have the potential to be an assistant diagnostic method for evaluating rectal cancer.Part ? The Value of Radiomics Based on Magnetic Resonance T2 WI in Evaluating the Treatment Response to Neoadjuvant Therapy for Rectal CancerPurpose: To explore the value of high resolution T2WI-based radiomics in evaluating treatment response to neoadjuvant chemoradiotherapy(n CRT)in patients with rectal cancer.Methods: We retrospectively analyzed the patients with rectal cancer confirmed by postoperative pathology from June 2016 to June 2019 in our hospital.All patients underwent 3-Tesla high-resolution T2 WI before and after n CRT.The lesions were segmented to draw the outline of VOI(VOIbefore and VOIafter).The radiomics features were extracted from the VOIs,then the features were selected by LASSO and PCA method to best predict the tumor regression grade(TRG)and pathologic complete response(p CR).The samples are randomly divided into training set(80%)and test set(20%)for machine learning.Four classifier models of Logistic Regression(LR),Random Forest(RF),Decision Tree(DT)and K-nearest neighbor(KNN)were used to obtain the receiver operating characteristic(ROC)curves and calculate the area under the curve(AUC),sensitivity and specificity.Compare the difference of ROC curves with De Long test.The decision curve analysis(DCA)was used to judge the clinical benefit.Results: A total of 80 patients including 29 patients(36.25%)had TRG 0-1,51(63.75%)had TRG 2-3,15 patients(18.75%)had p CR and 65(81.25%)had non-p CR.There were 3 features related to the TRG prediction and 11 features related to the p CR were obtained by LASSO method.The top six and five new features were recombined and selected by using the PCA method.For TRG group the RF classifiers was better than the others(P< 0.05).For p CR group the KNN classifier model has better diagnostic performance than the others(P< 0.05).The DCA shows that the LASSO method is superior to the PCA method for clinical benefit.Conclusion: Our study demonstrated that radiomics could evaluate the treatment response to n CRT,have the potential to be used to assist clinical decision-making in treatment.