Predicting spatio-temporal radiofrequency ablation temperature using deep neural networks

Area of Science:

• Biomedical Engineering
• Computational Modeling
• Pain Management

Background:

• Radiofrequency ablation (RFA) is a common treatment for facet joint pain, but it can cause multifidus muscle denervation.
• Computational simulations can help design new RFA techniques to prevent muscle damage, but can be computationally expensive.
• Deep neural networks (DNNs) offer a potential alternative for predicting tissue temperature during RFA.

Purpose of the Study:

• To predict the spatial and temporal tissue temperature distributions during RFA using DNNs.
• To develop a computationally efficient method for RFA simulation.
• To aid in the design of novel RFA treatments for facet joint pain.

Main Methods:

• Finite element (FE) models were used to simulate tissue temperature distributions for various probe distances.
• Temperature data from FE simulations were used to train DNNs for predicting spatio-temporal temperature profiles.
• The trained DNNs were validated using a separate dataset from FE simulations.

Main Results:

• DNNs accurately predicted tissue temperature distributions during RFA with an error rate as low as 0.05%.
• The DNN approach achieved a 92% reduction in computation time compared to traditional FE simulations.
• The model demonstrated high efficacy in predicting temperature variations within the tissue.

Conclusions:

• DNNs provide a computationally efficient and accurate method for predicting tissue temperature during RFA.
• This approach can significantly accelerate the design and optimization of RFA techniques for facet joint pain.
• The proposed method holds promise for developing RFA treatments that minimize unintended muscle denervation.