Fig 1.
Representative Hematoxylin and Eosin images of different models used to induce fatty liver.
A. Normal liver in wild type mouse fed chow diet. B. Mild fatty liver in wild type mouse fed high fat diet. C. Moderate fatty liver in GFP-LC3 mouse fed alcohol. D. Severe fatty liver in GFP-LC3 mouse fed high-fat high-carbohydrate diet.
Fig 2.
White regions in mouse liver biopsy.
Fig 3.
Overall flow diagram of the approach to automated macrosteatosis identification and quantification.
Fig 4.
Graphic showing dark and white regions with point and polygon annotations.
The white region labeled using point annotation (represented using 'X') is used for extracting attributes. A polygon annotation is shown using black polygon. The white regions falling within the polygon are indicated by arrows and these are the regions used for attribute extraction for the given label.
Fig 5.
Overall flow diagram of the approach to automated microsteatosis identification.
Fig 6.
A scatter plot showing correlation between model-computed percentage steatosis for mouse biopsies with respect to the corresponding pathologist grade.
Fig 7.
Mixed line/bar chart showing the correlation between the computed percentage steatosis with the expert pathologist grade.
The pathologist grade for mouse samples is shown as a bar matching the left axis, and the computed percentage steatosis is shown as a line matching the right axis.
Table 1.
The confusion matrix and metrics for the model built to detect macrosteatosis.
Table 2.
The confusion matrix and metrics for the model built to detect microsteatosis.
Fig 8.
A graph which represents the different models generated and their respective metrics.
Fig 9.
Comparison of annotated macrosteatosis lesions by pathologist v/s our model.
The left slide contains annotations given by the pathologist (marked as a point in magenta). The right slide contains macro fat lesions detected by the model (marked as point in dark blue).
Fig 10.
Comparison of annotated microsteatosis lesions by pathologist v/s our model.
The left slide contains annotations given by the pathologist (marked as polygons in dark blue). The right slide contains micro fat lesions detected by the model (marked as polygons in green).
Fig 11.
Some misclassified instances of microsteatosis lesions.
The left slide contains no micro fat according to the pathologist. The right slide contains wrongly classified micro fat lesions detected by the model (marked as polygons in green).