Artificial intelligence in the estimation of fat content in liver biopsies for liver transplantation: a systematic review

Quantification of hepatic steatosis is an important measure for assessing the viability of donor livers for transplantation. Currently, this measurement is obtained by a pathologist’s visual inspection of a liver biopsy sample to estimate the level of disease progression. However, recent advancements in automated methods, particularly artificial intelligence (AI) programs, offer alternative approaches. This systematic review summarizes the available literature on the various AI programs that have been designed to quantify hepatic steatosis. The literature search was performed using PubMed, Embase, and MEDLINE, covering studies published from 2019 to 2024. The 14 included studies used AI to calculate hepatic steatosis from digitized biopsies, except one, which used gross intraoperative smartphone images of the entire liver and provided a measure of accuracy. The programs use either a form of machine learning or image analysis software to label and quantify areas of steatosis on images of liver biopsies, reporting measures such as percent macrosteatosis (%MaS), percent total steatosis, fat fraction, nonalcoholic fatty liver disease activity score, and steatosis proportionate area. Agreement between AI-generated and pathologist %MaS scores varied widely across studies, but pathologist scores tended to be considerably higher than AI scores. Multiple studies noted significant interobserver variability between pathologists. Two papers reported associations between AI steatosis scores and post-transplantation outcome measures. Current literature shows promise for the application of AI in quantifying hepatic steatosis. This development provides the prospect of a faster, more precise method of determining the level of steatosis, which would further optimize the use of livers for transplantation.