The Application Of Texture Analysis Based On Enhanced CT Image In Diagnosis And Preoperative Quantitative Evaluation Of Malignant Degree Of Pancreatic Carcinoma

Part Ⅰ The application of matrix texture analysis based on enhanced CT image in the diagnosis of pancreatic cancerObjective:To explore the ability of texture analysis based on enhanced CT image in the diagnosis of pancreatic cancer,so as to provide a new method for qualitative and quantitative diagnosis of pancreatic cancer.Methods:49 patients with pathology proven posterior pancreatic cancer were analyzed retrospectively.Region of interest(ROI)was chosen at axial CT images with maximum enhancement of lesion and normal pancreas,and texture analysis were performed using Mazda software.The subsets of optimized texture parameters were chosen with four different methods,i.e.Fisher coefficient,the probability of classification error and average correlation(POE+ACC),mutual information measure(MI)and combination of the above three methods(FPM),respectively.The parameters were analyzed statistically in these two groups.The texture parameters were analyzed statistically in these two groups.Receiver operating characterist(ROC)was established for parameters to compare diagnostic performance.Results:The software extracted 6 histogram parameters,20 gray,level co-occurrence matrix parameters,2 autoregressive model parameters and 2 wavelet parameters among 30 characteristic parameters.The autoregressive model parameters Tetal and Teta4 showed no statistically significant difference between the lesion group and the normal group(P<0.05),while the other parameters were statistically significant(P<0.05).The ROC curve results showed that the highest AUC value of perc.10%in the histogram parameters was 0.961,and its corresponding sensitivity and specificity were 89.80%and 93.88%,respectively.The AUC values of S(2,0)SumAverg,S(2,0)SumAverg and S(2,0)SumAverg are the highest among the parameters of the gray-level co-occurrence matrix,all of which are 0.960.The sensitivity and specificity of the three are the same,which are 89.80%and 95.92%respectively.Among the wavelet transform parameters,wavenll_s-1 has the highest AUC value of 0.945,and its corresponding sensitivity and specificity are 83.67%and 95.92%,respectively.Conclusion:The characteristic parameters of texture analysis based on CT enhanced images were different between lesion and normal pancreas,providing a new method for qualitative and quantitative diagnosis of pancreatic cancer.Part Ⅱ Feasibility study of evaluation of malignant degree of pancreatic carcinoma by texture analysis based on CT contrast enhancementObjective:To explore the feasibility of texture analysis based on CT contrast enhancement in preoperative evaluation of the degree of malignancy of pancreatic cancer,and to evaluate the relationship between histogram parameters and pathological grade of pancreatic cancer.Methods:49 patients with pathology proven posterior pancreatic cancer including higher differentiation(27 cases)and lower-moderation differentiation(22 cases)were analyzed retrospectively.Region of interest(ROI)was chosen at axial CT images with maximum enhancement of lesion and texture analysis were performed using Mazda software.The subsets of optimized texture parameters were chosen with four different methods,i.e.Fisher coefficient,the probability of classification error and average correlation(POE+ACC),mutual information measure(MI)and combination of the above three methods(FPM),respectively.The parameters were analyzed statistically in these two groups.Receiver operating characterist(ROC)was established for parameters to compare diagnostic performance.In addition,the correlation was tested between parameters and groups.Besides,Raw data analysis(RDA),principal component analysis(PCA),linear discriminant analysis(LDA)and nonlinear discriminant analysis(NDA)were performed for texture classification.Results:FPM combined with NDA had the lowest misdiagnosis rate of 8.16%(4/49)in the evaluation of differentiation of pancreatic cancer.In the texture feature selection method,the error rate of FPM(8.16%-38.78%)was lower than MI(10.20%-38.78%),Fisher(14.29%-38.78%)and POE+ACC(14.29%-44.90%).In the classification method of texture features,the misdiagnosis rate(8.16%-14.29%)of NDA in assessing the degree of differentiation of pancreatic cancer was lower than that of LDA(12.24%-46.94%),PCA(34.69%-44.90%),and RDA(38.78%-44.90%),Among the best characteristic parameters extracted by Mazda software,mean value,50%Perc.,Teta2,and WavEnLL s-1 showed statistically significant difference between the two groups(P<0.05).Mean value,perc.50%,WavEnLL_s-1 were positively correlated with the degree of differentiation(r values were 0.02,0.02,0.03,all P<0.05),and Teta2 was negatively correlated with the degree of differentiation(r values were 0.03,P<0.05).The results of ROC curve showed that the AUC values of mean,perc.50%,Teta2 and WavEnLL s-1 were 0.695,0.694,0.699 and 0.684,respectively,and their sensitivity and specificity were 86.40%and 44.40%,respectively.86.40%,44.40%;90.91%,48.15%;Teta2 had the best diagnostic efficiency.Conclusion:The characteristic parameters of texture analysis based on CT enhanced images were different between higher differentiation and lower-moderation differentiation of pancreatic cancer,providing a new method for preoperative evaluation of the degree of malignancy of pancreatic cancer.