Real-Time Prediction of Coupled Electric and Temperature Fields in Radiofrequency Ablation: A Physics-Integrated...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Computational Physics

Background:

• Radiofrequency ablation (RFA) is a key minimally invasive treatment for solid tumors.
• Real-time monitoring of thermoelectric effects during RFA is crucial for treatment precision.
• Existing simulation methods are computationally intensive, hindering real-time clinical application.

Purpose of the Study:

• To develop a physics-integrated neural network model for real-time prediction of coupled electric and temperature fields during RFA.
• To overcome the computational limitations of traditional RFA simulations.
• To enable precise, personalized RFA treatments through rapid feedback.

Main Methods:

• A hybrid deep learning model combining DeepONet for electrical potential and ConvLSTM for temperature prediction was developed.
• The model was trained using data from thermoelectric coupling Finite Element Method (FEM) simulations.
• Model performance was validated using bio-mimic phantom experiments.

Main Results:

• The DeepONet model achieved high accuracy with MAE of 0.0241 and MRE of 1.44%.
• The ConvLSTM model demonstrated strong performance with MAE of 0.0286, MRE of 3.46%, and Dice score of 0.9334 for critical temperature zones.
• The developed model predicts RFA processes in under 1 second, significantly faster than traditional methods.

Conclusions:

• The physics-integrated neural network model offers rapid and accurate prediction of RFA thermoelectric effects.
• This AI-driven approach facilitates real-time monitoring and precise temperature control during RFA.
• Future integration with control algorithms can enhance 3D temperature field acquisition and treatment personalization.