Automated lipid droplets recognition in human steatotic liver : some preliminary results

The assessment of the degree of steatosis in routine liver biopsies represents an important task in different clinical situations, such as alcoholic steatohepatitis, non-alcoholic fatty liver disease, viral hepatitis, and evaluation of the viability of the graft in liver transplantation. Despite the advances in imaging techniques, microscopic examination remains the gold standard for the assessment of hepatic steatosis. In this study, we developed an automated approach for hepatic steatosis assessment in routine liver biopsies stained with Hematoxylin-Eosin (HE) from patients affected by hepatitis C. We performed a multi-step procedure by using a clustering technique, a two-levels thresholding and three shape parameters solidity, elongation and roughness to correctly distinguish fat droplets from other not stained objects like sinusoids. Lastly, we validated our results comparing them with those obtained by a pathologist via stereological point counting. We found a high agreement in the results, with a detection power of 91.01% and a false positive ratio of 4.49%.