A Preliminary Study on Fourier Neural Operator-based Surrogate Models for Radiofrequency Ablation for Small Liver Cancer Treatment Planning

Radiofrequency Ablation is a medical procedure that targets and destroys cancerous cells by heating tumours. This technique requires precise control to minimize damage to healthy tissues, which is often guided by computer-generated simulations. Traditional simulation methods are computationally intensive, taking hours to generate a simulation for a procedure that only lasts a few minutes. This study introduces the use of Fourier Neural Operators for simulating 2D temperature distributions of RFA over a liver region as a faster and reliable alternative that reduces simulation generation time by at least 80%, from 15 minutes to approximately 68 seconds with the predictions having marginal error, serving as a proof of con-cept for 3D. The effectiveness of FNO is demonstrated across various input resolutions and parameter settings of a 2D liver region, underscoring its robustness and adaptability, which could significantly improve preoperative planning and the efficacy of RFA treatments by decreasing the turnaround time between simulations and treatment.