INTRODUCTION Liver fibrosis is the accumulation of scar tissue caused by inflammation and cell death. Fibrotic liver is characterized by increased stiffness of the liver tissues, a feature that has been used by several methods to detect and stage liver fibrosis[1]. Recently, a number of studies have shown that liver deformation induced by the heart motion can be used to detect liver fibrosis. In these studies, the liver tissue deformation was achieved by ECG-gated tagging-based MRI techniques [2]. For example, Chung et al showed that there is a significant difference in the peak strain and displacement values among normal and cirrhotic (late stage of fibrosis) patients. In this work, we introduce a new set of features, namely, the histogram of the peak strain (HOPS) values within the liver that can be used to detect liver fibrosis. Results on 34 subjects showed potential of the HOPS as an indicator of liver fibrosis disease. Method and Materials ECG-gated tagged MRI was performed on 15 volunteers and 19 patients (with fibrosis stage from F1 to F3 diagnosed by Fibroscan and/or liver biopsy). Sagittal cross-sections (1-3 slices) were acquired with tag spacing=7mm, voxel size =1.18×1.18×8mm, 11-20 frame/cardiac cycle. Lagrangian strain tensors of all the points within the liver were calculated using phase-based tracking methods. Then, at each timeframe, the strain in the superior-inferior direction (major direction of motion), P1(t), and the strain in the perpendicular direction, P2(t), were computed. At each liver point, the peak tissue strain throughout the cardiac cycle is calculated to yield PP1 and PP2. The histogram of the peak strains within the liver is calculated for both directions to yield two feature vectors HOPS1 and HOPS2. Both vectors are also concatenated to form one feature vector, HOPS. Support Vector Machines classifier was used to classify the feature vectors. A leave-one-out cross-validation is used, by leaving one case as testing set and the remaining cases as training set. The process is repeated until testing all datasets. Results Figures 1 and 2 show the peak strain maps (PP1 and PP2) in a volunteer and a patient. Not only the peak strain value differs but also the distribution of the strain values. Table 1 summarizes the performance of the classifier when using the histogram of only one strain direction (HOPS1 or HOPS2) or both directions (HOPS) as feature vectors. The table, shows that the proposed method results in an accuracy of 85% when HOPS2 is used as a feature vector. conclusion A new method for detecting liver fibrosis using tagged MRI images was presented. The results show accuracy of 85% for patients with moderate fibrosis (stages from F1 to F3). [1] Cales P, Oberti F, Michalak S, et al. A novel panel of blood markers to assess the degree of liver fibrosis. HEPATOLOGY. 2005; 42:1373-1381. [2] Chung S, Mannelli B, and Axel L. et al., “Liver stiffness assessment by tagged MRI of cardiac induced liver motion,” Magn Reson Med., 65(4), 2011.

Acknowledgements

No acknowledgement found.

References

[1] Cales P, Oberti F, Michalak S, et al. A novel panel of blood markers to assess the degree of liver fibrosis. HEPATOLOGY. 2005; 42:1373-1381. [2] Chung S, Mannelli B, and Axel L. et al., “Liver stiffness assessment by tagged MRI of cardiac induced liver motion,” Magn Reson Med., 65(4), 2011