Quantitative magnetic resonance imaging of hepatic steatosis: Validation in ex vivo human livers

Abstract

Emerging magnetic resonance imaging (MRI) biomarkers of hepatic steatosis have demonstrated tremendous promise for accurate quantification of hepatic triglyceride concentration. These methods quantify the proton density fat-fraction (PDFF), which reflects the concentration of triglycerides in tissue. Previous in vivo studies have compared MRI-PDFF with histologic steatosis grading for assessment of hepatic steatosis. However, the correlation of MRI-PDFF with the underlying hepatic triglyceride content remained unknown. The aim of this ex vivo study was to validate the accuracy of MRI-PDFF as an imaging biomarker of hepatic steatosis. Using ex vivo human livers, we compared MRI-PDFF with magnetic resonance spectroscopy-PDFF (MRS-PDFF), biochemical triglyceride extraction, and histology as three independent reference standards. A secondary aim was to compare the precision of MRI-PDFF relative to biopsy for the quantification of hepatic steatosis. MRI-PDFF was prospectively performed at 1.5 Tesla in 13 explanted human livers. We performed colocalized paired evaluation of liver fat content in all nine Couinaud segments using single-voxel MRS-PDFF (n=117) and tissue wedges for biochemical triglyceride extraction (n=117), and five core biopsies performed in each segment for histologic grading (n=585). Accuracy of MRI-PDFF was assessed through linear regression with MRS-PDFF, triglyceride extraction, and histology. Intraobserver agreement, interobserver agreement, and repeatability of MRI-PDFF and histologic grading were assessed through Bland-Altman analyses. MRI-PDFF showed an excellent correlation with MRS-PDFF (r=0.984, confidence interval 0.978-0.989) and strong correlation with histology (r=0.850, confidence interval 0.791-0.894) and triglyceride extraction (r=0.871, confidence interval 0.818-0.909). Intraobserver agreement, interobserver agreement, and repeatability showed a significantly smaller variance for MRI-PDFF than for histologic steatosis grading (all P<0.001).

Conclusion: MRI-PDFF is an accurate, precise, and reader-independent noninvasive imaging biomarker of liver triglyceride content, capable of steatosis quantification over the entire liver.

© 2015 by the American Association for the Study of Liver Diseases.

Figures

Fig. 1. Experimental setup

(A) Photograph of an explanted liver with MR-visible markers labeling each liver segment for co-localization of MRI-PDFF measurements, MRS-PDFF voxels, core biopsies and tissue wedges. (B) High-resolution T1-weighted MR imaging allowed for positioning of MRS voxels ~1.5 cm below the markers as illustrated for liver segment II. (C) Five core biopsies (red lines) were obtained from each liver segment in a fan-shape sampling pattern in the same location using the surface markers for exact co-localization. (D) One co-localized tissue wedge (red-hatched wedge) was excised from each segment in the same location. Numbers indicate liver segments. Note the MR-visible markers on the MR images next to the indication of liver segments.

Fig. 2. Examples of a healthy liver vs. livers with moderate and severe steatosis

(A) Photographs demonstrate increased size and yellow hue with increasing steatosis (left to right). (B) MRI-PDFF enabled volumetric quantification of steatosis over the entire liver with exemplary results of 4.3%, 13.6% and 33.2% PDFF in segment VIII. (C) Histologic grading of co-localized core biopsies revealed that 8%, 30% and 75% of cells were affected by steatosis, respectively. Co-localized MRS-PDFF revealed fat-fractions of 4%, 14% and 33%, respectively. Triglyceride mass fractions from co-localized tissue wedges were 8%, 14% and 39%, respectively.

Fig. 3. Example of heterogeneous steatosis

(A) MRI-PDFF maps demonstrate highly heterogeneous steatosis, ranging from 10% in liver segment IVa to 20% in segment VII. Asterisks indicate MR visible markers. (B) Co-localized MRS-PDFF in all nine liver segments confirmed the inhomogeneous steatosis, ranging from 12% in segment IVa to 20% in segment VII. (C) Histologic grading from co-localized biopsies in each segment confirmed the inhomogeneous distribution of steatosis, ranging from 10% to 25% of cells affected by steatosis. Results of triglyceride analyses from co-localized tissue wedges for liver segments I to VIII were 8%, 10%, 13%, 7%, 7%, 11%, 11%, 17%, and 14%, respectively.

Fig. 4. Correlation of MRI-PDFF with MRS-PDFF, histological steatosis grade and extracted triglycerides

(A) MRI-PDFF showed an excellent correlation with MRS-PDFF (r=0.98, p<0.001). (B–C) MRI-PDFF showed a strong correlation with histological steatosis grading (r=0.85, p<0.001) and extracted triglycerides (r=0.87, p<0.001). Each of the 13 livers is indicated by nine symbols representing sampling from each of their 9 liver segments. Livers are sorted by increasing average steatosis grading.

Fig. 5. Inter- and intraobserver agreement of MRI-PDFF and histological steatosis grading

(A) Bland-Altman analyses of interobserver agreement of MRI-PDFF measurements and histological steatosis grading revealed a higher variance and larger bias of histological steatosis grading. (B) Results of intraobserver agreement also demonstrated a higher variance and larger bias of histological grading as compared to MRI-PDFF. Dotted lines indicate upper and lower 95% limit of agreement, solid line indicates bias. It is important to note that the bias and variance on the y-axis are expressed as percentages of the errors and not as absolute errors of the PDFF values and histologic steatosis grades, respectively. This representation has been chosen to allow of direct visual comparison of the variances and biases of the two techniques that have different metrics and different ranges. Note that due to this representation, the differences appear higher for smaller values of PDFF and histological grading than for larger values.

Fig. 6. Repeatability of MRI-PDFF and biopsy sampling

Bland-Altman analyses of (A) repeated MRI-PDFF measurements and (B) repeated histological steatosis grading revealed a higher interobserver variance and larger bias of histological steatosis grading. Dotted lines indicate upper and lower 95% limit of agreement, solid line indicates bias. It is important to note that the bias and variance on the y-axis are expressed as percentages of the errors and not as absolute errors of the PDFF values and histologic steatosis grades, respectively. This representation has been chosen to allow of direct visual comparison of the variances and biases of the two techniques that have different metrics and different ranges.

Fig. 7. Variability of histological fat assessment within segments and between segments

Five cores were obtained from each of the 117 liver segments in 13 livers, resulting in a total of 585 biopsies. Dot plots demonstrate the high variability of the 5 biopsies within individual liver segments (scatter in y-direction of a given liver segment). Moreover, dot plots indicate that actual differences in steatosis grade between individual liver segments exist (differences in y-direction between different segments). Dotted lines indicate the borders between steatosis grades: grade 0 (33%–66%), grade 3 (>66%). Solid line indicates averaged histopathologic steatosis degree in each liver. Red dots indicate biopsies that over- or underestimated the average