Value Of Multiparameter MRI Radiomics In Predicting Outcome Of Neoadjuvant Therapy For Mucinous Adenocarcinoma Of The Rectum

Objective: To investigate the correlation between the outcome of neoadjuvant therapy for rectal mucinous adenocarcinoma and feature texture labels extracted from multi-parameter MRI images.Methods: This study retrospectively collected the data of patients with rectal mucinous adenocarcinoma confirmed by postoperative pathology in Shengjing Hospital Affiliated to China Medical University from January 2014 to January 2016.According to the postoperative pathological and clinical tumor regression grading criteria(TRG 0-4 Rodel 2005),the outcomes of neoadjuvant therapy were divided into effective group(TRG 3-4)and ineffective group(TRG 0-2).MRI T1 WI,T2WI,DWI and enhanced T1 WI sequences were used as original images,and 3D-Slicer software was used to extract texture labels.A three-step feature selection method was used to screen all radiomics features for each sequence.Firstly,t test was used to roughly identify P & LT;Features of 0.1.Then,the minimum redundancy maximum correlation algorithm is applied to select 20 features with high correlation and low redundancy.Finally,the Akaike information criterion(AIC)was used to conduct the likelihood ratio test,and the reverse stepwise selection method was used to select the optimal feature subset for each sequence.The corresponding radiomics labels are constructed based on the optimal feature subset features of the Support vector machine(SVM)classifier.Multivariable linear regression analysis was used to select the optimal combination of radiomics tags from different sequences,and the radiomics nomogram was constructed.R software was used for feature selection and model construction.Results: ROC curve analysis showed that nomogram had better diagnostic performance than four single sequence labels in predicting the outcome of neoadjuvant therapy for mucinous adenocarcinoma of the rectum(AUC=0.975).Among the four single sequence tags,the DWI feature-based tags obtained higher AUC values than the T1 WI,T2WI and enhanced T1 WI tags(0.919 CI95% 0.836-0.968).Through DCA curve analysis,this study found that when the threshold probability changed from 0 to 1,the radiomics nomogram constructed based on T2 WI tags and DWI tags achieved the greatest net gain compared with the four single-sequence tags,the protocol assuming that all patients could achieve TRG grade 3-4 or no patients could achieve TRG grade 3-4.AUC can reach 0.975(CI95% 0.913-0.997).Conclusion: The feature texture labels extracted from multi-parameter MRI images have a high predictive value for the outcome of neoadjuvant therapy for rectal mucinous adenocarcinoma,and provide a reference for the treatment of patients with rectal mucinous adenocarcinoma.