Liver digital twins are computer models representing anatomy and physiology in an individualized, accurate and virtual way. In interventional radiology, these models could enable the simulation of several therapeutic strategies before the actual procedure. This is particularly valuable for Radiofrequency ablation (RFA), a well-established minimally invasive treatment for hepatic tumors, where predicting the induced ablation zone remains challenging. Since inaccurate predictions can lead to incomplete treatments or unintended damage to surrounding tissue, a liver digital twin, capable of precisely estimating the expected ablation volume and shape, could drastically improve the ablation outcome. Although such computational models of RFA have been proposed, they often depend heavily on manual processing of clinical data and are typically evaluated on a few carefully selected cases. In this paper, we present a retrospective clinical evaluation of a fully automatic patient-specific digital twin, capable of modeling the physical mechanisms involved in RFA of hepatocellular carcinoma (HCC). Our approach integrates clinical information extracted from CT images, from which level set representations of the liver anatomy and vasculatures are automatically extracted. The Lattice Boltzmann Method is used to estimate the temperature propagation and the resulting ablation zone subject to the underlying heat diffusion process. We evaluated our framework on a high-quality dataset comprising 26 patients. The results yield promising correlation between predicted and actual ablation volume (median Relative Volume Difference of 0.37 (IQ: 0.21, 0.51)) and highlight the superiority of our model compared to relying solely on the expected ablation zones as reported in the manufacturer’s instructions for use.