A machine learning surrogate model for fast approximation of simulated microwave ablation zones

Area of Science:

• Biomedical Engineering
• Computational Physics
• Medical Imaging

Background:

• Microwave ablation (MWA) is a minimally invasive treatment for various tumors.
• Finite element method (FEM) simulations accurately model MWA but are computationally intensive.
• Current simulation speeds limit interactive use in clinical treatment planning.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) surrogate model for rapid MWA ablation zone prediction.
• To approximate the 50°C isothermal contour, a key indicator of tissue ablation.
• To enable faster computational exploration for MWA treatment planning.

Main Methods:

• A feedforward neural network was trained using 2,625 FEM simulations.
• Input parameters included tumor geometry and dielectric properties.
• Model predictions were validated against FEM simulation results for accuracy.

Main Results:

• The ML surrogate model achieved rapid predictions of MWA ablation zones.
• High correlation (R > 0.95) was observed between ML predictions and FEM contours.
• Error analysis confirmed robust and accurate performance with distributions centered near zero.

Conclusions:

• The developed ML-FEM framework significantly accelerates MWA treatment planning by eliminating runtime simulations.
• This physics-informed approach offers a powerful tool for rapid computational exploration in MWA.
• ML-based surrogates show significant potential for advancing computational studies in medical ablation planning.