Predictive Value Of MRI T2WI Texture Analysis For Lymph Node Metastasis In Rectal Cancer

Objective: to extract the specific texture parameters of rectal cancer T2 WI images by texture analysis and to construct a comprehensive predictive model for preoperative diagnosis of rectal cancer lymph node metastasis.Materials and methods: The T2 WI images,serum tumor markers and basic clinical data of patients with rectal cancer who underwent rectal MRI scanning in Shengjing Hospital affiliated to China Medical University from January 2015 to December 2020 were analyzed retrospectively.Inclusion criteria:(1)rectal cancer was confirmed by pathology.(2)radical resection of rectal cancer and dissection of lymph nodes were performed without any radiotherapy and chemotherapy before MRI examination.(3)the pathological and related clinical data are complete.(4)there is no history of other malignant tumors.Exclusion criteria:(1)there are artifacts in the image,which affect observation or are difficult to delineate.(2)patients who did not undergo operation or only local resection could not obtain the pathological condition of lymph nodes.All patients received radical resection and lymph node dissection of rectal cancer in our hospital,and the pathological diagnosis was clear.The patients were randomly divided into training group and verification group according to the proportion of 7:3,which was used for training and verification of the prediction model.The artificial intelligence software was used to manually delineate the regions of interest(Region of Interest,ROI)of rectal cancer lesions and target lymph nodes on the T2 WI images of patients,and automatically extract the texture parameters of ROI,from which the texture feature parameters with statistical differences between lymph node metastasis group and non-metastasis group were selected.Logistic regression analysis was used to analyze the prediction models based on tumor tissue texture parameters,target lymph node texture parameters,patient clinical indicators and the combined data of the three.The area under the curve(Theareaunderthereceiveroperatingcharacteristiccurve,AUC)of the receiver working characteristic curve(Receiver operating characteristic curve,ROC curve)was used to evaluate the diagnostic efficacy of different models in identifying preoperative lymph node metastasis.Furthermore,De Long test was used to compare the AUC differences of each prediction model.Finally,the clinical benefit of each predictive model was evaluated by decision curve analysis(Decision Curve Analysis,DCA).P < 0.05 was taken as the standard of statistically significant difference.Results: Finally,112 cases were included in the study.Logistic regression analysis showed that the ratio of short diameter to long diameter of target lymph nodes(OR=40.503;95% CI:5.063-324.001)and the level of serum CA19-9(OR=1.044;95%CI:1.001-1.088)were independent predictors of lymph node metastasis in rectal cancer.In the clinical data prediction model training group,the AUC value(figure 3A)was 0.802(95% CI:0.696-0.884),the sensitivity was 70.005%,and the specificity was81.25%.In the verification group,the AUC value(figure 3B)was 0.696(95%CI:0.515-0.841),the sensitivity was 92.31%,and the specificity was 52.38%.401 texture feature parameters were extracted from the T2 WI images of tumor tissue and target lymph nodes.After screening,seven texture parameters from the tumor tissue and six from the target lymph nodes were retained.According to the target lymph node texture analysis prediction model,the AUC value was 0.881,sensitivity was86.67%,and specificity was 81.25% for the training group,and those for the verification group were 0.795,92.31% and 66.67%,respectively.According to the tumor tissue texture analysis prediction model,the AUC value of training group was0.844,the sensitivity was 80.005%,and the specificity was 79.17%.And in the verification group,the AUC value was 0.897,the sensitivity was 84.62%,and the specificity was 90.48%.Finally,the diagnostic efficacy of the joint predictive model combing texture parameters,target lymph node short / long diameter and patients’ serum CA19-9 level was significantly better than that of other models,and the difference was statistically significant.According to the joint predictive model,the AUC value of the training group was 0.978,the sensitivity and specificity were93.33% and 91.67%,respectively,and the AUC value of the verification group was0.897,the sensitivity was 84.62%,and the specificity was 90.48%.Conclusion: The comprehensive diagnostic model based on rectal T2 WI texture analysis combined with clinical indexes can effectively predict lymph node metastasis before operation,with potential clinical value.