Value Of Texture Analysis In Predicting Lymph Node Metastasis In Pancreatic Ductal Adenocarcinoma With Spectral CT Images-A Preliminary Study

Part 1 Texture analysis of spectral CT images in prediction of nodal involvement of pancreatic ductal adenocarcinoma——a feasibility studyObjective: To investigate the feasibility and influenced factors of texture analysis in predicting nodal involvement of pancreatic ductal adenocarcinoma with spectral CT images.Methods: 155 patients with surgery and pathology confirmed pancreatic ductal adenocarcinoma admitted to Ruijin Hospital were enrolled retrospectively.Gemstone spectral imaging(GSI)was performed in late arterial phase(AP,35s)and portal venous phase(PP,65s).Images were reconstructed with standard algorithm and slice thichness of 5mm and 1.25 mm.Study subjects were devided into two groups according to lymph node metastatic state:73 patients with lymph node metastases and 82 patients without nodal involvement.Texture features were extracted from manually drawn ROIs by using MaZda software.Features were selected according to the state of the lymph nodes.Feature selection methods including Fisher coefficient(Fisher),classification error probability combined with average correlation coefficients(POE+ACC),mutual information(MI),and combination of above three methods(FPM)were used.Four classification algorithms were used to predict the categories of the study cases by using the selected features,which include raw data analysis(RDA),principal component analysis(PCA),linear discriminant analysis(LDA),and nonlinear discriminant analysis(NDA).The results were presented by misclassification rate.Results:For the feature selection methods,FPM had the lowest misclassification rate.Among the classification algorithms,the misclassification rate of NDA was lower than those of RDA,PCA,and LDA.Among the four CT image series,the images of late arterial phase(AP)with 5mm slice thickness obtained the lowest misclassification rate(20.65%,32/155).Conclusions: Texture analysis of spectral CT images can be used as an assistant tool for predicting nodal involvement of pancreatic ductal adenocarcinoma.Feature selection with FPM and classification with NDA could obtain the lowest misclassification rate.Part 2 Resectable pancreatic ductal adenocarcinoma: association between preoperative texture features and nodal involvement in pathology findingsObjectives:To assess the value of texture features of the primary tumor from preoperative computed tomography images for predicting nodal involvement in pancreatic ductal adenocarcinoma.Methods: A total of 155 operable patients were enrolled in this retrospective study,who underwent contrast enhanced computed tomography spectral imaging,including late arterial phase and portal venous phase.There were 73 patients with lymph node metastases and 82 patients without nodal involvement.Texture analysis was performed by using MaZda software.A combination of feature selection algorithms(FPM)was used to determine 30 texture features with the discriminative power for differentiation between lymph node positive and negative groups.The prediction performance of the selected feature was evaluated by receiver operating characteristic curve analysis.Explore the relationship between these features and lymph node metastasis rate.Results: There were 10 significant texture features were identified,including 1 wavelet,2run-length matrix,and 7 co-occurrence matrix based features.They were WavEnLH_s-2,135dr_LngREmph and 135dr_Fraction of AP images,and 7 variables of sum average of PP images.Ideal cutoff value for predicting lymph node metastasis was 270 for WavEnLH_s-2(positive likelihood ratio 2.08).The 135dr_LngREmph value of positive lymph node metastasis group was higher than that of negative lymph node metastasis group;WavEnLH_s-2,135dr_Fraction and 7 mean values in lymph node metastasis positive group were all lower than those in negative group.In addition,135dr_LngREmph and 135dr_Fraction were correlated with the ratio of metastatic to examined lymph nodes.Conclusions: Preoperative computed tomography high order texture features are useful imaging signature for the prediction of nodal involvement in pancreatic cancer,and tumor with homogenous/uniformed density on late arterial phase and overall lower density on portal venous phase is more likely to have lymph node metastasis.