Purpose <p>Microwave ablation (MWA) is a locoregional, thermal destruction treatment approach for liver tumors. However, accurately predicting tissue necrosis remains challenging, primarily due to limitations in blood flow advection modeling, which is commonly approximated as volumetric perfusion, lacking personalization and clinical validation. This study evaluates the influence of perfusion modeling on MWA prediction accuracy in an in vivo study and proposes a patient-specific personalization framework yielding improved performance.</p> Methods <p>In vivo ablations were performed on porcine livers at sites with varying proximity to large vessels. Ablation zones were segmented from post-treatment CT scans for evaluating finite element simulations based on the Pennes bioheat equation. A two-compartment tissue vessel model was implemented, with proposed perfusion coefficients from in vivo calibration and a vessel radius model. The ablation extent and morphology were compared for the proposed approach and established perfusion values in the literature.</p> Results <p>The optimal liver perfusion coefficient varied markedly between experiments, confirming strong inter-subject variability. Personalizing the perfusion term improved Dice similarity by up to 55% and reduced Hausdorff distance by over 70% relative to literature values. Incorporating radius-dependent vessel perfusion produced further local improvements near major vessels, whereas variations in thermal and electrical properties yielded limited effects.</p> Conclusion <p>Personalized modeling of blood perfusion is essential for accurate MWA outcome prediction. The proposed framework shows that perfusion dominates uncertainty in thermal simulations and provides a pathway toward patient-specific digital twins for liver ablation planning.</p>