| Objective:To explore the value of texture analysis based on MRI T2WI in grading hepatocellular carcinoma.Methods:Clinical data of 134 patients with hepatocellular carcinoma were retrospectively analyzed.All the patients were confirmed by surgery and graded according to Edmondson-Steiner method.Mazda software was used to manually delineate the regions of interest in MRI T2WI images and extract and screen texture feature parameters.Mutual information(MI),Fisher coefficient(Fisher),classification error probability combined average correlation coefficient(classification error probability combined(?)verage correlation coefficients,POE+ACC)and three methods combined(FPM)were used to screen parameter.The feature classification methods adopted by the software include original data analysis(RDA),major component analysis(PCA),linear classification analysis(LDA)and non-linear classification analysis(NDA).And the results are presented with the misjudgment rate.Adopt the 30 textures screened by MPF method to analyze the texture feature in different grade hepatocellular carcinoma using logistic regression analysis,P<0.05 was considered statistically significantly.ROC curve was used to evaluate the discriminate efficiency of each parameter used alone or combined.Results:Among the texture feature parameter selection methods,the three methods of MI,Fisher coefficient and POE+ACC were close in grading hepatocellular carcinoma.In the classification analysis of texture features,the misdiagnosis rate of NDA(8.21%)in grading hepatocellular carcinoma were significantly lower than that of RDA,PCA and LDA,get the optimal diagnostic efficiency.S(0,5)Entropy,S(5,-5)Entropy,and WavEnHL_s2 were the independent factors in predicting high and low grade HCC.And the identification ability of combined three method three was improved compare with others.Conclusion:texture analysis based on MRI T2WI can be used to grading the pathological degree of hepatocellular carcinoma. |
