Comparative Analysis of PINN Architectures for Solving the Non-Dimensionalized Pennes' Bioheat Equation in...

Area of Science:

• Computational Biology
• Biomedical Engineering
• Thermal Physics

Background:

• Accurate heat transfer modeling in biological tissues is crucial for thermal therapies.
• The Pennes' bioheat equation is fundamental but challenging to solve in non-homogeneous tissues.

Purpose of the Study:

• To employ Physics-Informed Neural Networks (PINNs) for solving the non-dimensionalized Pennes' bioheat equation in non-homogeneous tissue.
• To enhance computational efficiency and stability through non-dimensionalization.

Main Methods:

• Utilized a custom PINN framework with NVIDIA Modulus to simulate the Pennes' bioheat equation.
• Implemented a non-dimensionalization process for spatial, temporal, and thermal parameters.
• Compared PINN performance against Finite Difference Method (FDM) solutions.

Main Results:

• PINNs, particularly Fourier-based architectures, demonstrated superior training stability and reduced loss.
• Non-dimensionalization improved the stability and efficiency of the simulations.
• PINNs effectively modeled heat transfer in non-homogeneous muscle and fat tissue variations.

Conclusions:

• PINNs combined with non-dimensionalization provide an effective and efficient computational tool for bioheat transfer modeling.
• This method advances the development of computational models for biomedical simulations and thermal therapies.
• The study highlights the potential of PINNs for optimizing clinical treatments like LITT and hyperthermia.