Automatic Liver and Subcutaneous Fat Segmentation from MRI-PDFF Images

Abdomen fat quantification and characterization is gaining importance in clinical settings for determining the concentration of different types of fats in various organs. However, due to a large amount of varying clinical data available, time is a significant constraint for radiologists. Therefore, this study is an attempt to develop a GUI based set of automated algorithms required by radiologists for quick and practical analysis. Methods: Different approaches for automatic segmentation and characterization of visceral and subcutaneous fat from MRI-PDFF abdomen images were evaluated and compared and later modified region growing algorithm was implemented for segmentation. Clinical correlation between subcutaneous fat volume and liver fat amount, between Body Mass Index and the liver fat amount and between BMI and subcutaneous fat volume was established. Results: Modified region growing algorithm was able to segment subcutaneous fat and liver from the MRI - PDFF images with the dice coefficient of 0.87 and 0.90, respectively. Clinical correlations were established and verified with literature. Conclusion: Modifications in region growing algorithm were successful with clinical relevancy.