Physics-Informed Machine Learning in Biomedical Science and Engineering.

Area of Science:

• Biomedical Science and Engineering
• Computational Biology
• Medical Physics

Background:

• Complex biomedical systems often defy traditional modeling due to data scarcity or intricate dynamics.
• Conventional black-box machine learning lacks the physical interpretability required for robust scientific insight.
• Physics-informed machine learning (PIML) offers a powerful alternative by integrating physical laws into data-driven approaches.

Purpose of the Study:

• To review and categorize major Physics-Informed Machine Learning (PIML) frameworks.
• To highlight the applications and potential of PIML in biomedical science and engineering.
• To identify current challenges and future research directions for PIML in the biomedical field.

Main Methods:

• Review of three primary PIML frameworks: Physics-Informed Neural Networks (PINNs), Neural Ordinary Differential Equations (NODEs), and Neural Operators (NOs).
• Discussion of how each framework embeds physical laws into machine learning models.
• Emphasis on applications in areas like biomechanics, medical imaging, and physiological systems.

Main Results:

• PINNs successfully model biosolid/biofluid mechanics, mechanobiology, and medical imaging.
• NODEs provide continuous-time modeling for dynamic physiological systems, pharmacokinetics, and cell signaling.
• Deep Neural Operators (NOs) enable efficient simulations across multiscale biological domains.

Conclusions:

• PIML frameworks like PINNs, NODEs, and NOs are transformative for biomedical modeling, especially when data is scarce or systems are complex.
• Key challenges include uncertainty quantification, generalization, and integrating PIML with other advanced AI, such as large language models.
• Advancing PIML promises more interpretable, accurate, and efficient solutions for complex biomedical problems.