Validation of fully-automated whole liver segmentation for measurement of hepatic fat fraction

Motivation: Hepatic fat fraction (FF) can be non-invasively measured using magnetic resonance imaging (MRI) to assess non-alcoholic fatty liver disease (NAFLD). Manual region of interest placement (ROI) is time-consuming and potentially inaccurate if liver lobes are not sampled evenly. Goal(s): The purpose of this work was to compare the FF obtained automatically and semi-automatically to those measured manually. Approach: Liver segmentations were automatically generated and manually corrected. FF was measured using manually placed ROIs in the liver, automatically from liver segmentation without corrections, and semi-automatically from manually corrected automatic liver segmentations. Results: FF from the three methods were strongly correlated and had excellent agreement. Impact: Hepatic fat fraction from MRI can non-invasively stage the degree of hepatic steatosis for NAFLD evaluation. Automatic fat fraction measurement is more efficient than manual approaches, making it more suitable for clinical workflows.