Recent advancements and applications of physics-informed machine learning in biomedical research

Abstract

Physics-informed machine learning (PIML) has emerged as a promising approach to modeling complex biomedical systems by integrating underlying biophysical laws with data-driven methods. Neural networks, particularly deep networks, are powerful data-driven function approximators that provide a flexible, scalable, and efficient framework for PIML, enabling the development of models with improved accuracy, interpretability, and robustness. This review examines recent advancements and applications of PIML in key biomedical domains where neural networks have been employed. We discuss core PIML techniques (i.e. physics-informed neural networks, constitutive artificial neural networks, and neural ordinary differential equations) for embedding physics into ML models and their applications in cardiology, oncology, radiology, and endocrinology, among other fields. By synthesizing recent progress and emerging applications from the scientific literature, we aim to highlight the potential of PIML in advancing both fundamental and translational research in biomedical engineering.