Machine Learning Improves MASLD Fibrosis Detection Over FIB-4

A research paper published on arXiv (2605.20523) investigates the use of machine-learning-enhanced non-invasive testing (MLE-NIT) for identifying advanced fibrosis in metabolic dysfunction-associated steatotic liver disease (MASLD). The study utilized three biopsy-validated groups from China, Malaysia, and India (n=784) to assess the performance of FIB-4 against a shallow-deep neural network (s-DNN), TabPFN, and large language models. The Chinese group, consisting of 486 for training and 54 for validation, was the basis for model development, while the final results were evaluated on cohorts from Malaysia and India. The models incorporated five factors: age, FIB-4, aspartate aminotransferase, platelet count, and alanine aminotransferase, with MLE-NIT aiming to enhance diagnostic precision while maintaining the FIB-4 variable framework.