Detection Of Liver Fibrosis Based On Intra Voxel Incoherent Motion Diffusion Weighted Imaging(IVIM-DWI)

Early detection of liver fibrosis is crucial to timely intervention and treatment and prevent further deterioration.The multiple b-value images of 45 subjects from Shenzhen Third People’s Hospital,including 26 healthy volunteers and 19 patients with liver fibrosis(4 of F0,5 of F1,3 of F2,7 of F3-4)were collected by using the intra voxel incoherent motion diffusion-weighted imaging(IVIM-DWI)technique.The double exponential parameter fitting method is applied to calculate the diffusion parameters such as Dfast,Dslow and Pf.Diffusion derived vessel density(DDVD)was also calculated using IVIM images with low b values.The four measures of different groups were then compared.It was found that the diffusion parameters of F0 group were higher than those of healthy volunteers.With the occurrence of liver fibrosis and deepening of fibrosis degree,the diffusion parameters decreased gradually,mainly due to the weakening of water molecular perfusion and diffusion effect in livers of patients with fibrosis.The ability of different parameters in distinguishing healthy volunteers from patients with different stages of fibrosis were further compared by scatter plots.It illustrated that the DDVD values combined with other diffusion parameters can improve the differentiation of different stages of fibrosis.This study also focuses on the automatic segmentation of liver ROI and extraction of image features for fibrosis detection.The U-net model was established to segment the liver region of right lobe and obtained Dice value of 0.9136.Then the Mazda software was used to extract features including the gray histogram,gray absolute gradient,run matrix,gray co-occurrence matrix,those acquired by the autoregressive model and wavelet transform,and the support vector machine classifier was adopted to detect fibrosis.The experimental results showed that the image features had better distinguishing power(accuracy of 0.9,AUC of 0.88-1.0,F1 of0.86-0.89)than the diffusion parameters(accuracy rate 0.7-0.8,AUC 0.38-0.75,F1 0.57-0.75).Moreover,automatic segmentation using U-net to extract imaging features is better than that of manually defining ROI in fibrosis recognition.This study proved that the diffusion parameters and image feature information based on IVIM-DWI technology could be effectively applied to determine liver fibrosis,and it had great potentials for the identification of early liver fibrosis.