Thermal damage map prediction during irreversible electroporation with U-Net

Area of Science:

• Medical Physics
• Oncology
• Artificial Intelligence in Medicine

Background:

• Irreversible electroporation (IRE) is an evolving cancer treatment.
• Accurate prediction of thermal damage is crucial for effective IRE treatment planning.
• Deep learning methods offer potential for enhancing treatment planning systems.

Purpose of the Study:

• To investigate the efficacy of the U-Net deep learning architecture for predicting thermal damage areas during irreversible electroporation (IRE).
• To develop a deep learning-based treatment planning system for IRE cancer therapy.

Main Methods:

• Liver tumor models with irregular shapes were created from MRI images using MIMICS and 3-Matic software.
• COMSOL Multiphysics 5.3 was used for finite element analysis to simulate electric field distribution and thermal damage.
• A U-Net deep learning network was designed and trained to predict thermal damage from electric field distribution data.
• Various electrode configurations, including pair needle, single bipolar, and multi-tine electrodes, were analyzed.

Main Results:

• The trained U-Net model achieved an average DICE coefficient of 0.96 and an accuracy of 0.98 for predicting thermal damage area.
• These high performance metrics were consistent across datasets including various electrode types.
• This marks the first application of the U-Net architecture for predicting thermal damage in IRE procedures.

Conclusions:

• The U-Net deep learning model demonstrates significant potential for accurately predicting thermal damage areas during irreversible electroporation.
• The findings support the use of U-Net as a component of a novel treatment planning system for IRE.
• This approach could enhance the precision and efficacy of IRE cancer treatments.