Automatic segmentation of liver/kidney area with double-layered fuzzy C-means and the utility of hepatorenal index for fatty liver severity classification.

BACKGROUND: Hepatorenal index (HRI) has been an efficient and simple quantified measure in distinction between normal and abnormalities of diagnosing fatty liver. However, considering the clinical significance, the diagnosis of severity stage is more important and single HRI cutoff may not be enough. Also, the segmentation of Liver/Kidney area should be automatic to get rid of operator subjectivity from ultrasonography analysis. METHOD: Double-layered Fuzzy C-Means (DFCM) pixel clustering method is proposed to extract the target area of analysis automatically. HRI and other shape related variables of Liver intensity distribution such as the skewness, the kurtosis, and the coefficient of variance (CV) are automatically computed for the fatty liver severity stage classification. RESULT: From fifty ultrasound images obtained from regular health checkup with 24 normal, 12 mild, 11 moderate, 3 severe stage determined by three different radiologists, the proposed DFCM automatically extracts the region of interests(ROI) and generates a set of statistically significant variables including HRI, the skewness, the kurtosis, the coefficient of variance of liver intensity distribution as well as liver echogenicity. In severity stage classification, the echogenicity of the liver and distribution shape variables such as the skewness and the kurtosis are better predictors than HRI based on our simple decision tree learning analysis. CONCLUSION: For better diagnosis of fatty liver severity stages, we need better set of features than the single HRI cutoff. Better machine learning structures are necessary in this severity stage classification problem with automatic segmentation method proposed in this paper.