Hybrid learning/numerical framework for fast and robust electric field simulation in irreversible electroporation.

Area of Science:

• Medical Physics
• Computational Biology
• Oncology

Background:

• Irreversible electroporation (IRE) is a promising non-thermal ablation for deep tumors.
• Accurate electric field distribution is crucial for IRE efficacy.
• Current numerical solvers are too slow for real-time clinical use.

Purpose of the Study:

• Develop a rapid, reliable workflow for electric dose mapping in IRE.
• Enhance treatment precision and patient outcomes through patient-specific dosimetry.
• Enable real-time adaptation of IRE procedures.

Main Methods:

• Propose a hybrid framework combining neural networks and classical solvers.
• Utilize convolutional neural networks for rapid electric potential field approximation.
• Refine approximations with lightweight iterative numerical correction for physical consistency.

Main Results:

• The hybrid solver achieves a 15-fold acceleration in computation time under homogeneous conductivity.
• In non-homogeneous settings, the method offers twofold speedup with improved accuracy over conventional solvers.
• Demonstrated robustness to electrode configurations and tissue heterogeneity in 15 patient cases.

Conclusions:

• The framework enables near-real-time, patient-specific dosimetry for IRE.
• Addresses a key barrier to clinical adoption of numerical simulations in IRE.
• Supports adaptive and precise tumor ablation through accelerated and reliable computations.