Simulation and parameter optimization of the temperature field in mild moxibustion using physics-informed neural...

Area of Science:

• Biomedical Engineering
• Computational Physics
• Thermal Medicine

Background:

• Accurate prediction of transient tissue temperature fields is crucial for understanding the thermal response of mild moxibustion.
• Optimizing treatment parameters requires reliable methods for simulating heat transfer in tissues.

Purpose of the Study:

• To develop and validate a physics-informed neural network (PINN) model for simulating bioheat transfer during mild moxibustion.
• To evaluate the influence of key treatment parameters on tissue temperature response.

Main Methods:

• A two-dimensional axisymmetric PINN model was developed, incorporating the Pennes bioheat equation and relevant boundary conditions.
• Nine orthogonal experimental cases were designed to assess parameter effects, with results validated against COMSOL finite-element simulations.
• The model was trained to predict transient tissue temperature fields at a depth of 5 mm.

Main Results:

• The PINN model demonstrated stable convergence and achieved high accuracy compared to the COMSOL benchmark (MAE: 0.159 °C, RMSE: 0.299 °C).
• Parameter sensitivity analysis revealed moxa-stick diameter and moxibustion distance as the most influential factors.
• An optimal parameter combination was identified for mild moxibustion conditions.

Conclusions:

• The proposed PINN framework offers a computationally efficient and physics-constrained tool for temperature-field reconstruction in mild moxibustion.
• This approach facilitates preliminary screening of treatment parameters, aiding in the optimization of mild moxibustion therapies.
• Further experimental and clinical validation is recommended to confirm the model's applicability.