Can machine learning diagnose metabolic dysfunction-associated steatohepatitis and fibrosis?

Metabolic dysfunction-associated steatohepatitis (MASH) is a progressive liver disease that can lead to fibrosis, cirrhosis, and liver failure. The standard for diagnosis is a liver biopsy, which is invasive and carries risks. Machine learning (ML) and deep learning (DL) models are being developed as non-invasive tools to diagnose MASH and liver fibrosis using data from blood tests, imaging, or genetic information. Current research shows these models can perform well, but they are not yet accurate enough to fully replace biopsy.

What the research says

A 2025 systematic review and meta-analysis evaluated 35 studies on ML and DL models for diagnosing MASH and fibrosis ³. The pooled area under the receiver operating characteristic curve (AUROC) — a measure of diagnostic accuracy where 1.0 is perfect — was 0.833 for ML models and 0.841 for DL models in diagnosing MASH ³. This indicates good performance, but not perfect. For fibrosis, the models also showed promising results, though the review noted variability across studies ³. Another study developed a metabolome-derived score using machine learning that identified MASH with an AUROC of 0.87 in a Chinese cohort and 0.81 in a Finnish cohort ⁹. This score also predicted liver-related mortality in large population studies ⁹. Additionally, a separate analysis used a neural network model based on gene expression to distinguish simple steatosis from MASH, achieving high accuracy in validation cohorts ⁶. While these results are encouraging, the meta-analysis highlighted that many models have not been validated in diverse populations, and their real-world performance may differ ³. The placebo response in MASH trials is also notable: about 11% of patients in placebo groups achieve MASH resolution without worsening fibrosis, which can complicate the assessment of diagnostic tools ⁴.

What to ask your doctor

• Are there any non-invasive tests or machine learning-based tools available at your clinic to help diagnose MASH or liver fibrosis?
• How accurate are these non-invasive methods compared to liver biopsy for my specific situation?
• If I undergo a machine learning-based diagnostic test, what would the results mean for my treatment plan?
• Should I consider a liver biopsy if non-invasive tests are inconclusive or suggest advanced disease?
• Are there any ongoing clinical trials at this center testing new machine learning diagnostic tools for MASH?