High-Throughput, Machine Learning–Based Quantification of Steatosis, Inflammation, Ballooning, and Fibrosis in Biopsies From Patients With Nonalcoholic Fatty Liver Disease

Abstract

Background & Aims Liver biopsy is the reference standard for staging and grading nonalcoholic fatty liver disease (NAFLD), but histologic scoring systems are semiquantitative with marked interobserver and intraobserver variation. We used machine learning to develop fully automated software for quantification of steatosis, inflammation, ballooning, and fibrosis in biopsy specimens from patients with NAFLD and validated the technology in a separate group of patients. Methods We collected data from 246 consecutive patients with biopsy-proven NAFLD and followed up in London from January 2010 through December 2016. Biopsy specimens from the first 100 patients were used to derive the algorithm and biopsy specimens from the following 146 were used to validate it. Biopsy specimens were scored independently by pathologists using the Nonalcoholic Steatohepatitis Clinical Research Network criteria and digitalized. Areas of steatosis, inflammation, ballooning, and fibrosis were annotated on biopsy specimens by 2 hepatobiliary histopathologists to facilitate machine learning. Images of biopsies from the derivation and validation sets then were analyzed by the algorithm to compute percentages of fat, inflammation, ballooning, and fibrosis, as well as the collagen proportionate area, and compared with findings from pathologists’ manual annotations and conventional scoring systems. Results In the derivation group, results from manual annotation and the software had an interclass correlation coefficient (ICC) of 0.97 for steatosis (95% CI, 0.95–0.99; P < .001); ICC of 0.96 for inflammation (95% CI, 0.9–0.98; P < .001); ICC of 0.94 for ballooning (95% CI, 0.87–0.98; P < .001); and ICC of 0.92 for fibrosis (95% CI, 0.88–0.96; P = .001). Percentages of fat, inflammation, ballooning, and the collagen proportionate area from the derivation group were confirmed in the validation cohort. The software identified histologic features of NAFLD with levels of interobserver and intraobserver agreement ranging from 0.95 to 0.99; this value was higher than that of semiquantitative scoring systems, which ranged from 0.58 to 0.88. In a subgroup of paired liver biopsy specimens, quantitative analysis was more sensitive in detecting differences compared with the nonalcoholic steatohepatitis Clinical Research Network scoring system. Conclusions We used machine learning to develop software to rapidly and objectively analyze liver biopsy specimens for histologic features of NAFLD. The results from the software correlate with those from histopathologists, with high levels of interobserver and intraobserver 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.

NAFLD: Time to apply quantitation in liver biopsies as endpoints in clinical trials

“…Unsurprisingly, the actual collagen content has been shown to increase exponentially, not linearly with fibrosis stage. 3 As such, the regression of 1 fibrosis stage from stage 4 to stage 3 would reflect a markedly higher anti-fibrotic effect than the regression from stage 2 to stage 1. In these cases, CPA may offer a more granular assessment, detecting small changes otherwise missed when comparing fibrosis stages.…”

“…4 Specifically, we have previously demonstrated that quantitation is more sensitive than the CRN scoring system in showing changes in a subgroup of liver biopsies. 3 This finding has been particularly striking in the assessment of inflammation and ballooning, both essential features for diagnosing NASH, and for which quantitation provides more subjective results. As suchthe ability of quantitation to define key histological features as continuous variables may delineate more subtle changes associated with a trial drug in NAFLD than is currently afforded by discrete scoring systems such as NASH CRN.…”

“…We agree with the conclusion of the investigators of this paperalong with other specialist groups 2on the need for different approaches, such as the use of quantitation, which provides more objective and reproducible results. 3 Following this, our group has recently developed a high-throughput, fully automated quantitation based on machine-learning. This methodology estimates the percentage of fat, inflammation, ballooning and fibrosis (collagen proportionate area, CPA) in routine histological images of liver biopsies from patients with NAFLD, with overall higher intraand interobserver variability than conventional histological assessment.…”

“…This methodology estimates the percentage of fat, inflammation, ballooning and fibrosis (collagen proportionate area, CPA) in routine histological images of liver biopsies from patients with NAFLD, with overall higher intraand interobserver variability than conventional histological assessment. 3 Several clinical trials have reported high screening failure rates due to low agreement between histopathologists. In the EMMINENCE study, the 3 expert pathologists agreed only on one-third of the screened patients meeting histological criteria.…”

Updates in the quantitative assessment of liver fibrosis for nonalcoholic fatty liver disease: Histological perspective

“…Furthermore, some clinical trials currently use a reduction of fibrosis histological stage by 1 or more as an outcome; however, a reduction in stage from 4 to 3 may have different implications regarding the efficacy of an antifibrotic drug as compared to a reduction from stage 2 to 1, given that CPA increases exponentially by fibrosis stage. 70 As a caveat though, sample adequacy becomes important when dealing with such precise measurements; moreover, adequate sample size differs between etiologies of cirrhosis. 71 CPA measurement is also still subject to technical issues such as variances in staining procedure, operator experience, and imaging software used.…”