Authors - Athanasios Angelakis, Gabriele De Vito, Eleni-Myrto Trifylli, Filomena Ferrucci Abstract - Advanced fibrosis is a major determinant of liver-related morbidity in metabolic dysfunction-associated steatotic liver disease (MASLD). FIB-4 is widely used as a first-line non-invasive test (NIT), but its fixed formula may underuse non-linear diagnostic information contained in age, aspartate aminotransferase (AST), alanine aminotransferase (ALT), and platelet count (PLT). We evaluated whether machine-learning-enhanced NITs (MLE-NITs) can improve advanced fibrosis detection while preserving the clinically accessible FIB-4 variable space. We used three biopsy-validated MASLD cohorts from China, Malaysia, and India (n = 784). The Chinese cohort was split into 486 training and 54 internal validation/tuning patients; final performance was reported only on the Malaysian (n = 147) and Indian (n = 97) external cohorts. Models used five variables: age, FIB-4, AST, PLT, and ALT. We compared FIB-4 with a shallow-deep neural network (s-DNN), TabPFN, and gpt-4o-2024-08-06 in zero-shot and fine-tuned settings. FIB-4 achieved external thresholded ROC-AUCs of 0.75 and 0.60 in Malaysia and India, respectively. TabPFN achieved 0.69 and 0.66, fine-tuned GPT-4o achieved 0.75 and 0.63, and the s-DNN achieved 0.77 and 0.67. The s-DNN contained only 354 trainable parameters, compared with 7,244,554 parameters for TabPFN, and provided the most balanced fixed-threshold operating profile. External diagnostics showed s-DNN Brier scores of 0.18 and 0.22, with AST and FIB-4 as dominant permutation-importance variables. Exploratory decision-curve analysis showed cohort-dependent clinical utility, favoring TabPFN in Malaysia and s-DNN in India.
Open Zoom