3269
A new technique for automatic removal of local errors in shear modulus estimation from measurement area in liver MR elastography1Office of Radiation Technology, Keio University Hospital, Shinjuku-ku, Tokyo, Japan, 2Department of Radiological Sciences, Graduate School of Human Health Sciences, Tokyo Metropolitan University, Arakawa-ku, Tokyo, Japan, 3Health Research Institute, National Institute of Advanced Industrial Science and Technology, Tsukuba-shi, Ibaraki, Japan, 4Department of Radiology, Keio University School of Medicine, Shinjuku-ku, Tokyo, Japan
Synopsis
Keywords: Data Processing, Elastography
Motivation: The stiffness value in two-dimensional liver MR elastography (MRE) is measured manually to avoid local errors in the stiffness estimation (dark/hot spots). Observer variability with this manual measurement is one of the obstacles to the clinical usage of MRE.
Goal(s): Our goal was to automatically remove dark/hot spots from the measurement area.
Approach: We introduced a new automated technique (coherent-wave auto-selection: CHASE) for measuring the stiffness value and tested it for the liver of five healthy volunteers.
Results: CHASE automatically generated a measurement area scarcely including dark/hot spots, resulting in high uniformity within that area.
Impact: The combination of our new technique and confidence mapping (clinical method) can reduce the process of manual measurement of stiffness value in liver MR elastography, resulting in improved variability and diagnostic performance for fibrosis staging.
INTRODUCTION
Magnetic resonance elastography (MRE) is a non-invasive MR-based technique that can estimate the mechanical properties of tissues and is one of the most accurate imaging modalities available for the staging of liver fibrosis 1-3. In the current clinical setting, liver MRE utilizes standardized two-dimensional (2D) MRE acquisition and analysis techniques by the quantitative imaging biomarker alliance (QIBA) 4. The liver MRE profile by the QIBA recommends that the shear modulus should be measured by manually avoiding areas of incoherent waves caused by wave interferences. This is because the areas of incoherent waves are subject to errors that locally decrease or increase the shear modulus, where these areas are called dark or hot spots 5. To improve the observer variability attributable to this manual measurement of the shear modulus, this study presents a new technique (called coherent-wave auto-selection: CHASE) for automating the removal of dark/hot spots from the measurement area 6. In this study, the performance of CHASE in liver MRE was evaluated by comparing shear modulus measurements with and without CHASE.METHODS
Five healthy volunteers were enrolled in this study. MRE experiments of the liver were performed on an MR imager (Discover 750 3.0 T, GE Healthcare, Waukesha, WI, USA) equipped with a 32-channel body array coil. Four slices of axial MRE images of the liver were acquired using a spin-echo-type echo-planar sequence at a typical mechanical vibration setup of 60 Hz, with imaging parameters recommended by the QIBA 4. CHASE automatically extract coherent wave areas where waves propagate in one direction (i.e., areas without wave interferences) by estimating the direction of wave propagation. In this study, CHASE processing was performed on the wave images created on the operating console using the parameters of the previous study 6. The liver shear modulus was estimated by using a multimodel direct inversion algorithm. As shown in Figure 1, the shear modulus values of the liver were measured with two regions of interest (ROIs): 1) the right lobe of the liver (right-liver ROI), 2) the area generated by CHASE (CHASE ROI), where their ROIs were set from the area removed the left lobe, major blood vessels, edge of the liver and cross-hatching marks (areas less than 0.95 on the confidence map). The distributions of the measured values of the shear modulus were analyzed with histograms of pixels of all volunteers within each ROI. The histograms were normalized to the number of pixels. Moreover, the difference in the mean value of the liver shear modulus in each ROI was assessed using a paired samples t-test (a total of 20 measurements were obtained for the liver [five volunteers, four slices per volunteer]). Statistical significance was set to P < 0.05.RESULTS
Examples of MRE images of the liver overlaid with the right-liver and CHASE ROIs are shown in Figure 2. Dark/hot spots were included within the right-liver ROI but not within the CHASE ROI. The histograms of the shear modulus of all volunteers for the right-liver and CHASE ROIs are shown in Figure 3. The relative numbers of pixels in lower shear modulus (0–1.5 kPa) were little difference between the right-liver and CHASE ROIs. In higher shear modulus (3.5–9.5 kPa), the relative numbers of pixels within the right-liver ROI were larger than those within the CHASE ROI, and thus the spread of the data in the CHASE ROI was narrower than that in the right-liver ROI. The mean value of the shear modulus within the CHASE ROI was significantly lower than that within the right-liver ROI (p < 0.01).DISCUSSION
The narrow histogram of the CHASE ROI shows that CHASE worked well to remove dark/hot spots from the measurement area, resulting in high uniformity within the CHASE ROI. Previous studies have reported that the shear modulus corresponding to fibrosis score F0 was less than 3.0 kPa 2,3. Considering these reports, the histogram result for the right-liver ROI indicates that hot spots affect the measurement of shear modulus compared to dark spots. In the CHASE ROI, as the large effects of hot spots were eliminated, the mean shear modulus within the CHASE ROI may be lower than that within the right-liver ROI.CONCLUSION
The results of this study demonstrate that CHASE can automatically remove dark/hot spots from the measurement area for the liver. This fact indicates that the process of manual ROI adjustments in liver MRE analysis can be reduced by combining the techniques of CHASE and confidence mapping (clinical method). CHASE is a useful technique that improves the variability of measurement of shear modulus and diagnostic performance of MRE.Acknowledgements
This work was supported by JSPS KAKENHI (Grant Number JP21K17548 and JP22K09338).References
1. Muthupillai R, Lomas DJ, Rossman PJ, et al. Magnetic resonance elastography by direct visualization of propagating acoustic strain waves. Science. 1995;269(5232):1854–1857.
2. Hsu C, Caussy C, Imajo K, et al. Magnetic Resonance vs Transient Elastography Analysis of Patients With Nonalcoholic Fatty Liver Disease: A Systematic Review and Pooled Analysis of Individual Participants. Clin Gastroenterol Hepatol. 2019;17(4):630–637.
3. Ozturk A, Olson MC, Samir AE, et al. Liver fibrosis assessment: MR and US elastography. Abdom Radiol (NY). 2022;47(9): 3037–3050.
4. MR Elastography Biomarker Committee. Magnetic resonance elastography of the liver, quantitative imaging biomarker alliance. Profile stage: Technically Confirmed. QIBA; February 14, 2022 Available from https://qibawiki.rsna.org/index.php/Profiles.
5. Yoshimitsu K, Shinagawa Y, Mitsufuji T, et al. Preliminary Comparison of Multi-scale and Multi-model Direct Inversion Algorithms for 3T MR Elastography. Magn Reson Med Sci. 2017;16(1):73–77.
6. Ito D, Numano T, Habe T, et al. A novel technique for automating stiffness measurement and emphasizing the main wave: Coherent-wave auto-selection (CHASE). Magn Reson Imaging. 2022;85:133–140.
Figures
