Photo Credit: Natali_Mis

An AI measurement tool delivered accurate and reproducible scoring measures in MASH, including fibrosis and progression-free survival.

AI-based measurement tool for scoring metabolic dysfunction-associated steatohepatitis (MASH) histology (AIM-MASH) predicted MASH Clinical Research Network necroinflammation grades and fibrosis stages that aligned with expert pathologist consensus scores, according to a study published in Nature Medicine.

Researchers developed AIM-MASH to assist in evaluating critical histologic signatures of MASH for clinical trials.

“Pathologist assessment of liver histopathology is central to the evaluation of disease severity and serves as the basis for patient selection and treatment efficacy assessment in MASH clinical trials. Histologic evaluation in MASH clinical trials has been limited by intra- and inter-pathologist variability in histologic grading and staging,” wrote corresponding author Andrew H. Beck, MD, PhD, and colleagues.

Although the FDA and the MASH Clinical Research Network have proposed panel scoring, guidance has lacked broad adoption. Consequently, variability in assessment remains high.

“This lack of reliability can reduce the power of MASH trials to detect a significant drug effect,” researchers wrote, “as trials are not typically powered to adequately account for such scoring variability.”

According to the study results, testing showed the AI measurement tool provided consistently accurate and reproducible scoring.

“The AIM-MASH versus consensus agreements were comparable to average pathologists for MASH Clinical Research Network scores (82% versus 81%) and fibrosis (97% versus 96%),” researchers reported.

AIM-MASH also provided continuous scores for key MASH histological features aligned with mean pathologist scores and noninvasive biomarkers. The continuous scoring system strongly predicted progression-free survival in patients with stage 3 and 4 fibrosis.

The study found that AIM-MASH achieved a high level of scoring accuracy and superior reproducibility compared with pathologists in retrospective assessment of endpoints in two clinical trials (EMMINENCE and ATLAS). Moreover, AIM-MASH detected a greater proportion of treatment responders than manual scoring, a finding observed in retrospective analyses of other trials.

“The results presented here highlight how collaboration between AI developers and pathologists with expertise in MASH can make consequential steps toward solving the problems inherent to MASH histologic assessment that lead to the failure of clinical trials,” researchers wrote. “To this end, AIM-MASH is being evaluated by both the FDA and the EMA [European Medicines Agency] for qualification as a drug discovery tool for use in clinical trials.”