The Role Of MRI Texture Analysis In Predicting NCRT Efficacy, TN Stage, And Neurovascular Invasion In Rectal Cancer

Objective:To investigate the potential of texture analysis of ADC map and T2WI to predict TN stage,neurovascular invasion and pCR after NCRT.Material and methods:79 patients with rectal adenocarcinoma who underwent surgery without NCRT(67 with ADC images,71 with T2WI)and 89 patients who went through NCRT before surgery(76 with ADC map,77 with T2WI)from Peking Union Medical College Hospital were enrolled.TexRAD software was applied for texture analysis.Set the spatial scaling factor value 0,2,3,4,5,6 to filter the image.Draw the ROI mannually along the tumor border on the image slice with the largest tumor cross section,and avoid obvious necrosis and mucus area.Six parameters(mean,sd,entropy,mpp,skewness,kurtosis)were calculated automatically.Postoperative pathology was set as gold standard.The parameters of ADC map and T2WI were compared between pTl-2 and pT3-4 stages,between pN0 and pNl-2,between neurovascular invasive and non-invasive group,between pCR and non-pCR group with Mann-whitney U test.The ROC curve was used to evaluate the diagnostic efficacy,and the independent predictors were selected by multivariate logistic regression analysis.Intraobserver variability of texture features extraction was evaluated by using intraclass correlation coefficients and bland-altman plot.Results:Mean、sd、entropy extracted from two sets of ROIs showed good or excellent agreement.There were significant differences in ADC-mean2,ADC-entropy,T2WI-sd0/4 and T2WI-entropy between stage pTl-2 and stage pT3-4.Multivariate logistic regression analysis suggested that ADC-entropy0,T2WI,entropy3(ADC-entropy0:AUC0.718,sensitivity 0.744,specificity 0.68.T2WI-entropy3:AUC0.706,sensitivity 0.854,specificity 0.533)are independent predictors of stage T3-4.The tumor transverse diameter was related to stage T3-4.Combined diagnostic performance of ADC-entropy0 and transverse diameter increased to 0.84,while T2WI-entropy3 and MRI-T stage,o.876.ADC-mean0/4 were significantly different between pNl-2 group and pN0 group,multivariate regression analysis showed mean0/4(ADC-mean0:AUC0.664,sensitivity 0.619,specificity 0.674;ADC-mean4:AUC0.673,sensitivity 0.667,specificity 0.744)were independent predictors of lymph node metastasis,and there was a correlation between MRI-T stage and lymph node metastasis.Combined diagnostic performance was 0.75.There were significant differences in ADC-sd2 and ADC-mean2-6 between the neurovascular invasive group and non-invasive group.Mean2A16 were independent predictors(mean2:AUC0.735,sensitivity 0.632,specificity 0.837;mean6:AUC0.686,sensitivity 0.474,specificity 0.918).Greater transverse,smaller vertical diameter,female,younger ages are associated with neurovascular invasion.Combined diagnostic performance was 0.89.ADC-mean4/5 and T2WI-sd6 achieved significant differences between pCR group and non-pCR group(P=0.03～0.04).Multivariate logistic regression analysis showed ADC-mean4 and T2WI-sd6(ADC-mean4:AUC0.669,sensitivity 0.833,specificity 0.534.T2WI-sd6:AUC0.65,sensitivity 0.818,specificity 0.564)are independent predictors of pCR.Advanced primary MRI-T stage and female are risk factors of non-pCR.Using a logistic regression model that incorporated ADC-mean4 and primary MRI-T stage or T2WI-sd6 and sex attributes to differentiate pCR from non-pCR,a moderate accuracy(AUC0.72-0.725)were achieved,respectively.Conclusion:MRI texture parameters derived from MRI images have potential to differentiate TN staging,neurovascular invasion and predict neoadjuvant therapy effect of rectal cancer.