Medical-AI trial: deep learning on repeated blood tests to catch liver fibrosis early (NIMIT-AI) — triage in fatty liver disease (MASLD) (NCT07675525)

Fatty liver disease (MASLD, metabolic dysfunction-associated steatotic liver disease) is a common condition in which fat accumulates in the liver and, over time, causes scarring (fibrosis). Finding scarring early makes it easier to act before it worsens. The tool commonly used now is the FIB-4 score computed from blood tests, but it misses many patients and cannot be calculated when lab values are incomplete.

The NIMIT-AI program that this study evaluates takes a different angle. Per the registry summary, rather than a single set of values, NIMIT-AI reads blood-test results across multiple visits to capture patterns suggesting fibrosis.

It was tested on 969 patients seen at Siriraj Hospital in Bangkok, Thailand between 2018 and 2022, and the primary endpoint is the area under the ROC curve (AUROC) for identifying significant fibrosis (F>=2). According to the summary, in this testing NIMIT-AI identified fibrosis more accurately than FIB-4 and continued to work even when some lab values were missing (missing values occur often in real practice).

What stands out is the use of a temporal trajectory as the cue and the reported robustness to the missing data that is inherent to real care. Capturing changes that a single-timepoint score can miss, from the stream of lab values that accumulates with each visit, points toward AI that leverages the electronic health record (EHR).

That said, this is a single-center, retrospective validation, and the reported advantage is within that scope. It does not establish clinical use or replacement of FIB-4, and reproduction in external populations remains future work.