Spatial characterization of intrahepatic fat deposition across steatotic liver disease subcategories: a multicenter study

Figures

• 

  Fig. 1. A schematic overview illustrating participant enrollment, and the inclusion and exclusion criteria. The initial case group included 12,477 patients diagnosed with fatty liver by abdominal MRI from three hospital centers. After screening using various exclusion criteria, the final analysis included 682 patients who were divided into the metabolic dysfunction-associated steatotic liver disease (MASLD) group, alcohol-related liver disease (ALD) group, and steatotic liver disease with chronic hepatitis B (SLD&CHB) group

• 

  Fig. 2. Schematic diagram of the study design. The nnU-Net model was trained using in-phase (IP) and opposed-phase (OP) images as two-channel input. The liver parenchyma was delineated with avoidance of the gallbladder and major vessels (e.g., inferior vena cava and main portal vein), while intrahepatic small vessels were retained. Meanwhile, the subdivisions (left, right, and caudate lobes) were also integrated into framework for concurrent delineation and segmentation training. The left and right lobes were defined based on the midline connecting the inferior vena cava and middle hepatic vein, with the caudate lobe delineated separately. Model-generated masks of the liver and its subdivisions (left [red mask], right [green mask], and caudate lobe [blue mask]) were overlaid onto the reconstructed fat and reconstructed water images, and were used for fat fraction analysis. The fat fraction was calculated at the voxel level using the given formula, and the fat fractions in the whole liver and its subdivisions (left and right lobes) were analyzed. To examine the fat spatial distribution relative to the portal vein, the liver volume was divided into 100 concentric spherical shells (yellow…

• 

  Fig. 3. Association between hepatic fat fraction and spatial heterogeneity across liver regions among SLD subcategories. Scatterplots with locally weighted regression (LOWESS) smoothing curves depict the relationship between hepatic fat fraction and the standard deviation of intrahepatic fat distribution (SD-FF), across three SLD subgroups. Each dot represents an individual patient. Extreme values (top and bottom 1%) were excluded to reduce the influence of outliers

• 

  Fig. 4. Distribution curves of fat fraction in the right and left lobes across SLD subcategories. Comparative density plots of the hepatic fat fraction in the right and left lobes. Bandwidth was optimized using Silcerman’s rule. The dotted line marks the mean fat fraction. Data were analyzed using R v4.0 (ggplot2 package)

• 

  Fig. 5. Periportal-to-peripheral spatial distribution of hepatic fat fraction among SLD subcategories at different steatosis grades (L1–L3). From the reference point of the bifurcation of the main portal vein, the liver volume was divided into 100 concentric spherical shells (radial zones) based on increasing distance from the portal origin. The mean fat fraction was calculated for each shell, to assess the regional variation in fat accumulation along the portal-to-peripheral axis. For each disease group at a specific distance, the bandwidth represents the fat fraction range, and the dot represents the median fat fraction. The trend line represents the periportal-to-peripheral distribution of the fat fraction