AI fatty liver detection is being tested as a way to find people whose liver disease has gone unnoticed in ordinary care. The pitch is less futuristic than it sounds: have software review blood work, health records and imaging that clinicians already collect, then flag patients who may need follow-up.
That matters because fatty liver disease often produces no clear symptoms while damage accumulates. Wired reports that it affects roughly 30% of adults worldwide. Fat buildup can trigger inflammation and scarring, known as fibrosis; progressive disease can lead to cirrhosis and liver failure, and is associated with higher cardiovascular-disease and cancer risk.
Early identification does not make an algorithm a treatment. It can, however, give clinicians a chance to assess a patient before advanced scarring appears. Wired reports that early damage may be reversible, while NYU Langone notes that the condition can remain silent until liver injury has already occurred.
How could AI detect fatty liver disease earlier?
The near-term use is triage. Jeffrey Lazarus of the CUNY Graduate School of Public Health and Health Policy told Wired that systems could retrospectively inspect large volumes of hospital visits and lab reports to prioritize people at greater risk. Risk factors cited in the reporting include obesity, high cholesterol and type 2 diabetes.
One practical target is FIB-4, a noninvasive score used to estimate the risk of advanced fibrosis. It combines a patient's age, two liver-enzyme measurements and a blood-clotting measure. The calculation is not exotic, which is precisely the point. Software could calculate it automatically from routine tests rather than leave another task in a crowded primary-care workflow.
FIB-4 has limits. Wired reports that it is less accurate for adolescents and older adults, and can generate false positives and unnecessary specialist referrals if it is not paired with further testing. In people with concerning liver fat, combining FIB-4 with an enhanced liver fibrosis blood test improved identification of advanced fibrosis four-fold, according to the report.
Imaging offers another possible signal. Researchers at Osaka Metropolitan University reported that a model reading routine chest x-rays identified fatty liver disease with 82% accuracy. Chest films are ordered to examine the heart and lungs, but can include part of the liver. The proposed workflow is to surface that incidental clue for a clinician, not to declare a diagnosis from an x-ray alone.
What evidence exists for AI liver-risk tools?
Several reported studies suggest machine-learning models may outperform basic scores in selected settings. Evido's LiverPRO uses age and nine routine blood biomarkers; Wired reports it outperformed FIB-4 at predicting serious liver-problem risk in more than 470,000 middle-aged people. Evido is commercializing the tool with Roche. Separately, an international group of hepatologists reported that the blood-test-based ALADDIN model outperformed FIB-4 and other risk scores when identifying patients most likely to benefit from resmetirom.
Those are promising comparisons, not proof that health systems can deploy the tools broadly and improve patient outcomes. A 2021 review in Clinical Liver Disease described potential AI roles in identifying patients and assessing disease severity, while Wired reported that liver-care use has largely remained in research.
Paul Brennan, a University of Dundee specialty registrar, told Wired that these systems would not replace imaging or biopsies. Their plausible job is the dull but consequential one: a smarter first pass that sends the right people for further evaluation while avoiding a flood of unnecessary referrals.
This story draws on original reporting from WIRED.
