Study On The Prediction Of Neurovascular Invasion In Colorectal Cancer Based On The Radiomics Features Of Extracellular Volume Parameter Map By Spectral CT

Objective To explore the imaging model based on Extracellular volume(ECV)of spectral CT,in order to predict perineural and vascular invasion(including blood vessels and lymphatic vessels)in Colorectal cancer(CRC)before surgery.Materials and Methods A total of 308 patients with CRC confirmed by pathology were retrospectively collected and randomly divided into a training set(n=216)and a testing set(n=92)at 7:3.Each patient underwent enhanced spectral CT examination before surgery.Five parameters(Mono E＿40ke V,Conventional,Effective Z,Electron Density,Iodine Density)of the spectral CT images were collected and used to calculate the ECV.After preprocessing the above images,ITK-Snap software was used for semi-automatic segmentation to delineate tumor 3D Volume of interest(VOI).Fe Ature Explorer(FAE)software was used to extract the features.Pearson Correlation Coefficients(PCC)were used to reduce the dimensions.Analysis of Variance(ANOVA),Relief algorithm(Relief),Recursive Feature Elimination(RFE)and Kruskal-Wallis test(KW)were used for feature selection.Finally,Support Vector Machine(SVM),Linear Discriminant Analysis(LDA),Logistic Regression via Lasso(LR),and Logistic Regression via Lasso(LRLasso)were modeled as classifiers,and single-parameter and multiparameter models were established respectively.The efficacy of ECV model and each parameter model was compared,radiomics score(Radscore)was calculated,and the clinical risk factors were combined with Logistic regression to construct a column graph.Finally,the efficacy of pure clinical model,radiomics model and combined model was compared.The area under the curve(AUC)of receiver operating characteristic curve(ROC),decision curve analysis(DCA)to evaluate the value of different models in predicting vascular,lymphatic and perineural invasion in colorectal cancer.Results The results of AUC in the groups of vascular invasion,lymphatic invasion and perineural invasion showed that the ECV model had higher efficacy than other single parameter models,and the efficacy of the multi-parameter model was comparable.The AUC of training set and testing set in vascular invasion group were ECV: 0.855,0.809;Mono E＿40ke V: 0.766,0.691;Conventional: 0.713,0.654;Effective Z: 0.771,0.771;Electron Density: 0.743,0.596;Iodine Density: 0.753,0.807;Multi＿parameter: 0.851,0.838.The AUC of training set and testing set in lymphatic vessel invasion group were ECV: 0.855,0.885;Mono E＿40ke V: 0.689,0.723;Conventional: 0.764,0.582;Effective Z: 0.738,0.730;Electron Density: 0.714,0.715;Iodine Density: 0.763,0.655;Multiparameter: 0.831,0.924.The AUC of training set and testing set in the perineural group were ECV: 0.835,0.892;Mono E＿40ke V: 0.701,0.631;Conventional:0.754,0.554;Effective Z: 0.805,0.705;Electron Density: 0.687,0.740;Iodine Density: 0.798,0.714;Multi＿parameter: 0.845,0.811.Logistic regression determined clinical indicators(T/N staging)as risk factors.Compared with pure clinical model and pure radiomics model,the combined model combined with ECV radiomics features improved the diagnostic efficiency.Compared with the AUC of pure clinical model,pure radiomics model and combined model,In the vascular invasion group,the training set and testing set AUC were Clinical: 0.658 and0.682;Radiomics: 0.855 and 0.809;Combined: 0.858 and 0.846.In the lymphatic vessel invasion group,the training set and testing set AUC were Clinical 0.684 and 0.704;Radiomics0.855 and 0.885;Combined 0.869 and 0.900.In the perineural group,the training set and testing set AUC were Clinical: 0.675,0.632;Radiomics: 0.835,0.892;Combined: 0.847,0.901.Conclusion The radiomics features based on extracellular volume of spectral CT can effectively predict the vascular,lymphatic and perineural invasion of colorectal cancer.Combined with clinical risk factors,it can improve the diagnostic efficiency and has the value of non-invasive preoperative diagnosis.