Physics-informed neural networks for physiological signal processing and modeling: a narrative review

Area of Science:

• Biomedical Engineering
• Computational Science
• Artificial Intelligence

Background:

• Physiological signal analysis traditionally relies on empirical models.
• Emerging techniques like Physics-Informed Neural Networks (PINNs) offer a novel approach.
• PINNs integrate physical laws into neural network training for enhanced performance.

Purpose of the Study:

• To provide a comprehensive overview of PINNs in physiological signal analysis.
• To categorize existing research based on key theoretical and practical aspects.
• To identify challenges and future research directions in this domain.

Main Methods:

• Systematic literature review and screening of relevant studies.
• Categorization of over 40 selected studies.
• Analysis of input data types, applications, physics-informed models, and neural network architectures.

Main Results:

• PINNs demonstrate significant advantages in generalizability, data efficiency, and interpretability for physiological signals.
• Successful application in solving forward and inverse problems within this domain.
• Identified key challenges including noisy inputs, computational cost, and model configuration.

Conclusions:

• PINNs offer a powerful framework for advancing physiological signal analysis.
• Addressing identified challenges is crucial for broader clinical adoption.
• This review serves as a guide for researchers, promoting precise and data-efficient solutions.