Predictive Value Of Multi-parameter Mri And Radiomics Analysis On The Main Pathological Features Of Rectal Cancer

Objective: To evaluate the value of multi-parameter MRI and radiomics in predicting the main pathological features of rectal cancer.Materials and Methods: In this study,prospective experimental scheme was adopted to collect patients with rectal cancer from July 2019 to December 2020.A total of 110 patients with rectal cancer were included in this study.All patients underwent radical surgery for rectal cancer without any preoperative chemoradiotherapy or endoscopic treatment.Clinical biochemical indexes such as fecal occults blood,CEA,CA199,CA125,CA153,CA242 and CA724 were recorded.Multi-parameter rectal MRI was performed two weeks before surgery,and the detailed pathological examinations of the postoperative specimen were performed,TNM staging was performed.The major pathological features of rectal cancer involved in this study included pathological type of rectal cancer（non-mucinous adenocarcinoma or mucinous carcinoma）,degree of differentiation,depth of invasion,epithelial-mesenchymal transformation（expression of E-cadherin and Vimentin）,and expression of aquaporin-1.This study was mainly divided into three parts.The first part discussed the clinical characteristics of patients with rectal cancer and the predictive value of conventional MRI features（lesion size,location,T2 signal intensity ratio,circumferential margin,invasion of extramural blood vessels,invasion of extramural fat,etc.）for the main pathological features of rectal cancer.In the second part,we discussed the predictive value of MRI diffusion models（including diffusion weighted imaging--DWI,in voxels incoherent motion--IVIM,diffusion kurtosis imaging--DKI）and quantitative parameters of dynamic enhancement（Ktrans,Kep,Ve values）for the main pathological features of rectal cancer.The third part was to used machine learning method to analyze MRI radiomics characteristics,and establish prediction model based on clinical,image data and radiomics characteristics to evaluate the main pathological characteristics of rectal cancer.The main statistical methods in this study included the following: Independent sample t test or Mann-Whitney U,one-way analysis of variance（one-way ANOVA）or Kruskal-Wallis H test,χ2 or Fisher’s exact test,Spearman correlation analysis,receiver operating characteristic curve（ROC）analysis,the predicted diagnostic efficacy was described by area under curve（AUC）,and P < 0.05 was considered statistically significant.Results: A total of 110 patients with rectal cancer were included in this study,including 65 males and 45 females,age 60.7±12.5 years（range 34-89 years）.Clinical stage can be divided intoⅠstage 23.6%,Ⅱ stage 37.3%,Ⅲ stage 35.5%,Ⅳ stage 3.6%;T staging was mainly T3（52.7%）,the N and M stages were mainly N0M0（62.7%,96.4%）.The results of ROC analysis showed that:（1）Patients’ age and CA724 had diagnostic efficacy for rectal mucinous carcinoma（AUC= 0.667 and 0.681,P < 0.05）,and CEA had diagnostic efficacy for the serosal layer and outside infiltration of rectal cancer（AUC=0.723,P < 0.05）,AFP had a mild diagnostic efficacy for low expression of E-cadherin（AUC=0.616,P=0.046）,all clinical indicators discussed in this study had no diagnostic efficacy for differentiation degree of rectal cancer,expression of Vimentin and Aquaporin-1（P > 0.05）;（2）In the conventional MRI features,T2 signal intensity ratio had a high diagnostic efficacy for rectal mucinous carcinoma（AUC=0.885,P < 0.05）,extramural fat invasion had diagnostic efficacy for the differentiation degree of rectal cancer（AUC=0.65,P < 0.05）,the length of the tumor,the distance from the lower margin of the tumor to the anus,and the invasion of extramural fat had diagnostic efficacy on the serosal layer and outside invasion of rectal cancer（AUC = 0.611,0.632,0.618,P < 0.05）,conventional MRI features had no diagnostic effect on E-cadherin expression（P>0.05）,T2 signal intensity had a mild diagnostic efficacy for high expression of Vimentin（AUC=0.619,P=0.037）,The T2 signal intensity ratio and the distance from the lower edge of the tumor to the anus had mild diagnostic efficacy for high expression of Aquaporin-1（AUC = 0.639 and 0.615,P < 0.05）;（3）ADC values（ADC₆00,ADC₁000,ADC₂000,and ADC₃000）with b values of 600,1000,2000,and 3000 s /mm2 had high diagnostic efficacy in the diagnosis of rectal mucous carcinoma,with AUC of 0.839,0.874,0.831,0.93,respectively（P < 0.05）,they had diagnostic efficacy for both serosal layer and outside infiltration of rectal cancer,with AUC values of 0.745,0.69,0.708 and 0.709,respectively（P < 0.05）,there was no diagnostic efficacy for the degree of differentiation of rectal cancer（P>0.05）,ADC₆00,ADC₂000 and ADC₃000 had diagnostic efficacy for low expression of E-cadherin（AUC = 0.628,0.63,0.63,P < 0.05）,ADC₃000 had diagnostic efficacy for high expression of Vimentin（AUC=0.644,P=0.011）,ADC₁000,ADC₂000 and ADC₃000 had diagnostic efficacy for the high expression of Aquaporin-1 in rectal cancer（AUC value was 0.619,0.660 and 0.715,P < 0.05）;（4）The quantitative parameter D value of IVIM-MRI had a high diagnostic efficacy for rectal mucinous carcinoma（AUC=0.876,P < 0.05）,D* value had diagnostic efficacy for the serosal layer and outside infiltration of rectal cancer（AUC=0.641,P < 0.05）,D value had diagnostic value for the high expression of Aquaporin-1（AUC=0.636,P < 0.05）,IVIM quantitative parameters had no diagnostic efficacy on differentiation degree,E-cadherin and Vimentin expression of rectal cancer（P > 0.05）;（5）Quantitative parameter Diffusivity and Kurtosis value of DKI-MRI had high diagnostic efficacy for rectal mucous carcinoma,with AUC of 0.876 and 0.9,respectively（P < 0.05）.The degree of differentiation,the depth of invasion,and the expression of E-cadherin and Vimentin in rectal cancer had no diagnostic efficacy（P > 0.05）,Diffusivity and Kurtosis value had diagnostic value for high expression of Aquaporin-1（AUC was 0.633,0627,P < 0.05）;（6）DCE-MRI quantitative parameters Ktrans and Kep values had diagnostic efficacy for rectal mucinous carcinoma（AUC was 0.683,0.673,P < 0.05）,both Ktrans and Ve values had diagnostic efficacy for highly differentiated rectal carcinoma（AUC = 0.655,0.65,P < 0.05）,and had diagnostic efficacy for the serosal layer and outside infiltration of rectal cancer(AUC = 0.785,0.785,P < 0.05,Ktrans and Kep values had diagnostic efficacy for low expression of E-cadherin（AUC = 0.801,0.722,P < 0.05）,and high expression of Vimentin（AUC = 0.724,0.628,P < 0.05）,Ktrans value had diagnostic efficacy for high expression of Aquaporin-1（AUC=0.632,P=0.018）.The results of rectal MRI radiomics analysis showed that LR classifier based on T2 WI sequence had the highest predictive power for rectal mucinous carcinoma（AUC=0.921）,and KNN classifier had the highest predictive power for highly differentiated rectal carcinoma（AUC=0.819）;The SVM classifier based on the dynamic enhanced axial sequence of venous phase had the highest predictive power for the serosal and outside infiltration of rectal cancer（AUC=0.854）,and the highest predictive power for the low expression of E-cadherin in rectal cancer（AUC=0.825）.The SVM classifier based on dynamically enhanced venous sagittal sequence had the highest predictive power for high expression of Vimentin（AUC=0.809）;The SVM classifier based on T2 WI sequence had the highest predictive effect on the high expression of Aquaporin-1 in rectal cancer（AUC=0.917）.conclusion: Clinical biochemical examination has limited value for predicting the main pathological features of rectal cancer.T2 WI signal intensity is more effective in the diagnosis of rectal mucinous carcinoma,while other conventional MRI features were less valuable in the diagnosis of the main pathological features of rectal cancer.The quantitative parameters of DWI（with ultra-high b value）,IVIM,DKI and DCE-MRI have certain diagnostic value for the main pathological features of rectal cancer.Moreover,the radiomics features significantly improved the predictive performance of multi-parameter MRI,suggesting that the radiomics features based on machine learning have the potential to be a noninvasive biomarker to evaluate the main pathological features of rectal cancer preoperatively.