TWIN-SCAN: Liver Biomarker Estimation Using Machine Learning and Digital Twin Simulation

Sumaiya Afroz Mila, Sandip Ray

Confirmation: I have read and agree with the IEEE BHI 2025 conference submission's policy on behalf of myself and my co-authors.

Keywords: Early-Stage Liver Disease, Biomarker Estimation, Digital Twin Modeling, Machine Learning, Bilirubin Conjugation

TL;DR: A comparative modeling analysis of estimated bilirubin values from an ML-based method and digital twin–based bilirubin conjugation for early-stage liver disease monitoring and disease progression or treatment effect tracking.

Abstract: Early-stage liver diseases often progress silently and remain undiagnosed due to the lack of labeled clinical biomarker data. This study presents a comparative modeling framework to predict stage-specific liver biomarker ranges using limited real-world data. We compare two complementary approaches: (1) a machine learning (ML) based mapping method trained on cirrhosis-stage data and healthy references, and (2) a physiology-informed partial digital twin (DT) model that simulates bilirubin conjugation across various stages of liver disease. The digital twin is further personalized using patient-level parameters such as age and gender to enhance accuracy. Results demonstrate that the digital twin approach yields biomarker predictions that are more biologically plausible and better aligned with clinical trends. This framework highlights the potential of physiological modeling to complement data-driven methods in generating accurate, non-invasive biomarker estimates for early-stage liver disease detection during data scarcity.

Track: 7. General Track

Registration Id: FQNLZMTJ5CM

Submission Number: 46