A Subvoxel Correction of Boiling‐Induced Susceptibility Artifacts in Magnetic Resonance Thermometry and Dosimetry for Monitoring Microwave Thermoablation: A Feasibility Study in a Swine Model

ABSTRACT Purpose Real‐time monitoring of microwave liver ablation (MWA) using MRI thermometry can be hindered by boiling‐induced susceptibility artifacts. These artifacts cause large temperature measurement errors that prevent accurate lesion size prediction. This study proposes a correction methodology based on removing the contribution of this susceptibility artifact using subvoxel sources of susceptibility. Methods In vivo microwave ablations ( N = 23) were performed on seven pig livers and the temperature was monitored using the proton resonance frequency shift (PRFS) method. The boiling‐induced artifacts were observed in 11 ablation cases. To validate the proposed methodology, the thermal dose was computed from the corrected temperature maps, and the resulting lesion estimates were compared in three dimensions with ground‐truth lesion segmentations derived from post‐ablation T1w images. A numerical simulation was also conducted to emphasize the need for a finer spatial discretization. Results After correction, a better agreement was noticed between thermal dose prediction and lesion size. Over the 11 cases observed with boiling‐induced susceptibility artifacts, the median volumetric Dice, Total Overlap, and False Negative rates changed by 4.8%, 6.2%, and −11.4%. Conclusion Comparison with T1w imaging showed improvements in prediction of lesion volume. The proposed methodology takes into account partial volume effects and the contribution of adjacent slices. It was able to simulate a diverse range of deformed dipole‐like artifacts observed in experimental data.