PhysRFANet: Physics-guided neural network for real-time prediction of thermal effect during radiofrequency ablation treatment

Abstract

Radiofrequency ablation (RFA) is a minimally invasive technique that is widely used to ablate solid tumors. Achieving precise personalized treatment requires feedback information on in situ thermal effects induced by RFA. Although computer simulations facilitate the prediction of electrical and thermal phenomena associated with RFA, their practical implementation in clinical settings is hindered by their high computational demands. In this paper, we propose a physics-guided radiofrequency ablation neural network (PhysRFANet) to enable real-time prediction of thermal effect during RFA treatment. Three networks, an encoder-decoder based convolutional neural network (EDCNN), U-Net, and attention U-Net, designed for predicting the temperature distribution and the corresponding ablation lesion, were trained using biophysical computational models that integrated electrostatics, bioheat transfer, and cell necrosis, along with magnetic resonance (MR) images of breast cancer patients. The computational model was validated through experiments using ex vivo bovine liver tissue. Our model demonstrated a Dice score of 96.3% in predicting lesion volume and a root mean squared error (RMSE) of 0.5624 for temperature distribution when tested with foreseen tumor images. Notably, even with unforeseen images, it achieved a Dice score of 93.8% for the ablation lesion and an RMSE of 0.7078 for the temperature distribution. All networks were capable of inferring results within 10 ms. The proposed technique, applied to optimize the placement of the electrode for a specific target region, holds significant promise for enhancing the safety and efficacy of RFA.