High-Throughput, Machine Learning–Based Quantification of Steatosis, Inflammation, Ballooning, and Fibrosis in Biopsies From Patients With Nonalcoholic Fatty Liver Disease
Roberta Forlano,Benjamin H. Mullish,Nikolaos Giannakeas,James B. Maurice,Napat Angkathunyakul,Josephine Lloyd,Alexandros T. Tzallas,Markos Tsipouras,Michael Yee,Mark R. Thursz,Robert D. Goldin,Pinelopi Manousou
DOI: https://doi.org/10.1016/j.cgh.2019.12.025
IF: 13.576
2020-08-01
Clinical Gastroenterology and Hepatology
Abstract:Background & AimsLiver biopsy is the reference standard for staging and grading non-alcoholic fatty liver disease (NAFLD), but histologic scoring systems are semi-quantitative with marked inter- and intra-observer variation. We used machine learning to develop fully automated software for quantification of steatosis, inflammation, ballooning, and fibrosis in biopsies from patients with NAFLD and validated the technology in a separate group of patients.MethodsWe collected data from 246 consecutive patients with biopsy-proven NAFLD and followed in London, the United Kingdom, from January 2010 through December 2016. Biopsies from the first 100 patients were used to derive the algorithm and biopsies from the following 146 were used to validate it. Biopsies were independently scored by pathologists using the nonalcoholic steatohepatitis clinical research network criteria and digitalized. Areas of steatosis, inflammation, ballooning, and fibrosis were annotated on biopsies by 2 hepatobiliary histopathologists to facilitate machine learning. Images of biopsies from the derivation and validation sets were then analyzed by the algorithm to compute percentages of fat, inflammation, ballooning, and fibrosis, as well as collagen proportionate area (CPA), and compared with findings from pathologists' manual annotations and conventional scoring systems.ResultsIn the derivation group, results from manual annotation and the software had an inter-class correlation coefficient (ICC) of 0.97 for steatosis (95%CI, 0.95–0.99; P<.001); ICC, 0.96 for inflammation (95%CI, 0.9–0.98; P<.001); ICC, 0.94 for ballooning (95%CI, 0.87–0.98; P<.001); and ICC, 0.92 for fibrosis (95%CI, 0.88–0.96; P=.001). Percentages of fat, inflammation, ballooning, and CPA from the derivation group were confirmed in the validation cohort. The software identified histological features of NAFLD with levels of inter- and intra-observer agreement ranging from 0.95 to 0.99; this value was higher than that of semi-quantitative scoring systems which ranged 0.58 to 0.88. In a subgroup of paired liver biopsies, quantitative analysis was more sensitive in detecting differences compared to the nonalcoholic steatohepatitis Clinical Research Network scoring system.ConclusionsWe used machine learning to develop software to rapidly and objectively analyse liver biopsies for histologic features of NAFLD. The results from the software correlate with those from histopathologists, with high levels of inter- and intra-observer agreement. Findings were validated in a separate group of patients. This tool might be used for objective assessment of response to therapy for NAFLD in practice and clinical trials.
gastroenterology & hepatology