MRI-Based Morphological Features And Deep Learning For Evaluation Of Local Staging In Stage T0-2 Rectal Cancers

Part Ⅰ Submucosal Enhancing Stripe as a Dynamic Contrast-Enhanced MRI-Based Imaging Feature for the Differentiation of Stage T0-1 from Early T2 Rectal CancersPurpose Accurate differentiation of stage T0-1 rectal tumors from stage T2 rectal tumors facilitates the selection of appropriate surgical treatment.Magnetic resonance imaging(MRI)is a recommended technique for local staging,but its ability to distinguish T1 from T2 tumors is poor.To explore the value of submucosal enhancing stripe(SES),an uninterrupted enhancing band between the rectal tumor and the muscular layer on dynamic contrast-enhanced T1-weighted imaging(T1WI),as a potential imaging feature for differentiating T0-1 from T2 stage rectal tumors.Materials and Methods In this retrospective study,patients with pT0-1 and pT2 stage rectal tumors who underwent pretreatment MRI and rectal tumor resection between January 2012 and November 2019 were included.Two radiologists independently evaluated tumor characteristics(SES,status of muscularis propria[SMP],tumor shape,location,and distance from tumor to anal verge,maximum tumor length,and maximum tumor circumference)on MRI.The associations of clinical and imaging characteristics with T0-1 or T2 stages were assessed with uni variable analysis;multivariable regression analysis with backward stepwise selection was applied to build the predictive model.The receiver operating characteristic(ROC)curve was plotted to evaluate the power of the model.The diagnostic accuracies for the differentiation of T0-1 from T2 tumors using SES and SMP were compared using the McNemar’s test.Results Data in 431 patients(mean age ± standard deviation,60 years ± 10;261 men)were evaluated.In total,249 had T0-1 lesions and 182 had T2 lesions.SES(β,3.9;95%confidence interval[CI]:3.1,4.7;P<0.001),SMP(β,1.3;95%CI:0.7,1.9;P<0.001),and carpet-like shape(β,1.6;95%CI:0.5,2.8;P=0.01)were independent factors distinguishing T0-1 tumors from T2.The multivariable model showed diagnostic ability,with an area under the receiver operating characteristic curve of 0.92(95%CI:0.90,0.95).The diagnostic accuracy was 87%(376 of 431;95%CI:84%,90%)for SES and 67%(290 of 431;95%CI:63%,72%)for SMP(P<0.001).Conclusions Submucosal enhancing stripe on contrasted-enhanced MRI,status of muscularis propria on T2-weighted images,and tumor shape can serve as independent imaging features for differentiating T0-1 from T2 rectal tumors.Moreover,submucosal enhancing stripe is a more accurate feature than status of muscularis propria.Part Ⅱ Prediction of Lymph Node Metastasis in Stage T1-2 Rectal Cancers with MRI-Based Deep LearningPurpose Accurate lymph node metastasis(LNM)prediction in patients with stage T1-2 rectal cancer is essential for choosing appropriate treatment options but remains a clinical challenge.This study aimed to investigate whether a deep learning(DL)model based on preoperative MR images of primary tumors can predict lymph node metastasis(LNM)in patients with stage T1-2 rectal cancer.Materials and Methods In this retrospective study,patients with stage T1-2 rectal cancer who underwent preoperative MRI between October 2013 and March 2021 were included.All patients underwent direct surgical resection and were assigned to the training,validation,and test sets.Lesion Annotation was performed by two gastrointestinal radiologists.Four two-dimensional and three-dimensional(3D)residual networks(ResNet18,ResNet50,ResNet101,and ResNet152)were trained and tested on T2-weighted images to identify patients with LNM.Three radiologists independently assessed LN status on MRI,and diagnostic outcomes were compared with the DL model.Predictive performance was assessed with AUC and compared using the Delong method.Results In total,611 patients(mean age ± standard deviation,59 years±11;356 men)were evaluated(444 training,81 validation,86 test).The AUCs of the eight DL models ranged from 0.80(95%confidence interval[CI]:0.75,0.85)to 0.89(95%CI:0.85,0.92)in the training set and from 0.77(95%CI:0.62,0.92)to 0.89(95%CI:0.76,1.00)in the validation set.In the test set,the AUCs ranged from 0.61(95%CI:0.46,0.76)to 0.79(95%CI:0.70,0.89).The ResNet101 model based on 3D network architecture achieved the best performance in predicting LNM in the test set,with an AUC of 0.79(95%CI:0.70,0.89)that was significantly greater than that of the pooled readers(AUC,0.54[95%CI:0.48,0.60];P<0.001).Conclusion The DL model based on preoperative MR images of primary tumors outperformed radiologists in predicting LNM in patients with stage T1-2 rectal cancer.