Optimization of regions of interest sampling strategies for proton density fat-fraction MRI of hepatic steatosis before liver transplantation in ex vivo livers

Abstract

Objectives: The quantity of regions of interest (ROIs) constitutes the primary determinant of the time investment in image analysis. In the context of proton density fat-fraction (PDFF) magnetic resonance imaging (MRI) conducted on liver grafts in ex vivo conditions, this research systematically examines various ROI sampling strategies. The findings of this study furnish essential insights, offering a foundation for optimizing time efficiency while ensuring precise assessment of hepatic steatosis before the crucial process of liver transplantation.

Methods: This was a retrospective analysis of a prospective study and included 35 liver grafts with histopathological steatosis that underwent 3T PDFF MRI in ex vivo. One ROI of 1 cm² was selected for each hepatic segment, and any combination of ROIs in 1-8 liver segments was used, resulting in 511 combinations. Using intraclass correlation coefficients (ICCs) and Bland-Altman analyses, the PDFFs of all these combinations were compared with the 9-ROI average PDFF. There was a moderate correlation between the average PDFF and the histological findings (R = 0.47, P＜0.01).

Results: The average 9-ROI PDFF of all liver grafts was 4.07 ± 4.35 % (0.870-20.904). All strategies with ≥5 ROIs had intraclass correlation coefficient (ICC) ≥ 0.995 and absolute limits of agreement (|LOA|)≤ 1.5 %. Overall, 54 of 84 (67.5 %) 3-ROI sampling strategy had ICC ≥0.995, and 70 of 84 (70 %) had |LOA|≤ 1.5 %. A total of 111 of 126 (88.1 %) 4-ROI sampling strategy had ICC ≥0.995, and 125 of 126 (99.2 %) had |LOA| ≤ 1.5 %.

Conclusions: The employment of the 5-ROI sampling strategy proves instrumental in both time conservation and precise assessment of hepatic steatosis within liver grafts during the ex vivo phase preceding liver transplantation.

Keywords: Hepatic steatosis; Liver transplantation; Proton density fat fraction; Sampling strategy.

Fig. 1

Flowchart shows process of selecting liver grafts.

Fig. 2

Representative images of an 18-year-old man liver graft. (a–c) Pathology confirmed as macrovesicular hepatic steatosis (hematoxylin and eosin; original magnification,×100, × 200 and × 400, respectively); (d–f) Oil red staining confirmed as macrovesicular hepatic steatosis (original magnification,×100, × 200 and × 400, respectively); (g) Visual inspection; (h) ROI of the liver on FatFrac image of IDEAL-IQ; (i) ROI of the liver on T2-weighted image.

Fig. 3

Bland-Altman plot of the left and right lobe proton density fat-fraction values in the 35 liver grafts of this study.

Fig. 4

The intraclass correlation coefficients (ICCs) (y-axis) for the different sampling strategies in relation to the number of ROIs (y-axis). Each dot represents a particular sampling strategy (511 in total). The ICC increased with the number of ROIs.

Fig. 5

Level of agreement (LOA) bounds (y-axis) shown for the different sampling strategies by the number of regions of interests (ROIs). Each dot represents a particular sampling strategy (511 in total). The LOA decreased as the number of ROIs per strategy increased.