| Objective: To utilize computer-assisted diagnostic method to identify early cirrhosis and advanced cirrhosis, and discuss the value of multiple sequences MRI with assistance of computer in the diagnosis and identification of cirrhosis.Method: We chose 186 patients(male 120 cases, female 66 cases) who had accept MRI enhancement scan on upper abdominal, whose average age is(49±12.0). Their medical records were divided into cirrhosis group(n = 148) and normal group(n = 38). Then the cirrhosis group was divided into early cirrhosis group(n=53) and advanced cirrhosis(n=95). Firstly,(1)we compared early cirrhosis group with advanced cirrhosis through MRI of 5 different sequences and manually selected ROI(region of interest); The size of the 30*30, The total number of 2675, 526 ROI of normal group, 814 ROI of early cirrhosis and 1335 ROI of advanced cirrhosis were selected in this study.(2)We extracted 56 texture features of 14 characteristics parameters(including relevant information of 1, relevant information of 2, the normalized deficit, the normalized inverse difference moment, ed.) in four directions(0°, 45°, 90°and 135°) through Gray-level Co-occurrence Matrix.(3)We utilized the BP neural network classifier with ten times cross validation method to identify normal liver, early cirrhosis and advanced cirrhosis. then,(1)Selection of portal venous phase images. We extracted 80 texture features of 20 characteristics parameters(including prominent clustering, dark clustering, energy, homogeneity reliability, entropy ed.) in four directions(0°, 45°, 90°and 135°) through Gray-level Co-occurrence Matrix.(2)Through the box chart method for feature selection.(3)We utilized the neural network classifier to identify early cirrhosis and normal liver, normal liver and advanced cirrhosis, early cirrhosis and advanced cirrhosis by pairwise classification.Results: The result of 56 texture feature showed that the dynamic enhancement in portal vein phase had the highest accurate rate and the most obvious effect of identification of early cirrhosis, advanced cirrhosis and normal liver tissue, and its correct rate was 87.68%(555 / 633), while the T2 WI classification results showed the worst, the accurate rate was only 68.50%(261/381). The images effect display of equilibrium phase rated second, and the accuracy rate was 84.35%(528/626). The correct rate of T1 WI and arterial phase images for classification of early cirrhosis, advanced cirrhosis and normal liver were 71.82%(339/472), and 70.17%(374/533), so the difference between theses two were not obvious.In MRI of 5 sequences, T1 WI had the highest accuracy rate of identification of the normal liver tissue with the correct rate of 93.88(138/147); dynamic enhanced scanning of portal vein phase showed the best correct rate in early cirrhosis identification and the correct rate of 86.43%(172/199); dynamic enhanced scanning of equilibrium phase had the highest accuracy rate in advanced cirrhosis identification, whose correct rate was 92.31%(348/377).Based on 80 texture feature, the analysis of 52 texture feature of portal vein phase through feature selection showed that the correct rates of identification of normal liver and early cirrhosis, normal liver and advanced cirrhosis, early cirrhosis and advanced cirrhosis were 88.53%, 78.95%, 74.49%, which were all higher than those without feature selection(86.74%, 75.94%, 72.30%).Conclusion: Based on computer aided diagnosis of cirrhosis identification:(1)MR images of normal liver, early cirrhosis and advanced cirrhosis can be identified by neural network classifier with ten times cross validation method for texture feature. In computer-assisted multiple sequences MRI for cirrhosis identification, dynamic enhanced scanning portal veins phase images can be the first choice, followed by the equilibrium phase, T1 WI, arterial phase and T2 WI.(2)After feature selection, the texture feature can improve the accuracy rate of identification of normal liver, early cirrhosis and advanced cirrhosis. |
PDF Full Text Request