Physics-guided neural network for tissue optical properties estimation.

Area of Science:

• Biomedical optics
• Computational biology
• Medical imaging

Background:

• Accurate estimation of tissue optical properties is crucial for biomedical applications like diagnostics and therapies.
• Traditional methods rely on physics-based models or data-driven machine learning, each with limitations.
• Integrating physics with machine learning offers a promising avenue for enhanced prediction accuracy.

Purpose of the Study:

• To develop and evaluate a novel physics-guided neural network (PGNN) for tissue optical properties regression.
• To investigate the generalizability of PGNN compared to pure artificial neural network (ANN) models.
• To assess the performance of PGNN on both in-domain and out-of-domain datasets.

Main Methods:

• A physics-guided neural network (PGNN) was proposed, integrating physical priors and constraints into an artificial neural network (ANN).
• The model was trained and tested using simulated single-layered tissue samples generated via Monte Carlo simulations.
• Performance was evaluated using both in-domain and out-of-domain test datasets to assess generalizability.

Main Results:

• The proposed PGNN demonstrated superior generalizability compared to a pure ANN model.
• PGNN achieved better prediction accuracy and robustness across both in-domain and out-of-domain test datasets.
• The integration of physics principles significantly enhanced the network's ability to generalize.

Conclusions:

• Physics-guided neural networks offer a powerful approach for accurate and generalizable estimation of tissue optical properties.
• This hybrid methodology overcomes limitations of purely physics-based or data-driven methods in bioimaging.
• PGNN holds significant potential for advancing biomedical diagnostic and therapeutic applications.