337 - Modeling Temperature Changes in a Cryoablation Phantom Using Spectral CT

Purpose: Over the past decade, spectral CT–based thermometry has attracted growing interest as a technique for intraprocedural monitoring during cryoablation, particularly for assessment of the critical –20 °C to –40 °C “kill zone” {1,2}. To further evaluate and characterize performance, we developed a low-cost phantom model to test the feasibility of spectral CT thermometry within temperature ranges relevant to cryoablation.

Materials and Methods: A tissue phantom was created using compacted ground beef, into which three temperature probes were inserted. The phantom was cooled in a -80 °C freezer for 48 hours to simulate cryoablation. Subsequently, the phantom was placed in a spectral CT scanner and allowed to thaw while continuous temperature and intermittent CT data was acquired.

Regions of interest were manually segmented at probe tips (containing 533, 604 and 598 voxels). Conventional HU (CHU), Electron Density (ED), and Physical Density (PD) values were obtained across 20 temperature points ranging from –48 °C to –11 °C {3}.

Probe data was combined via an “unpooled” and “pooled” method. Unpooled analysis treated all probe acquisitions separately (n = 60 bins; Group A). Pooled analysis was performed by pooling all voxel data corresponding to n = 20 (Group B) and n=15 (Group C) equally distributed temperature bins ranging from –44 to – 11 °C.

Linear regression was performed on mean voxel values. Noise was estimated using the signal range-to-average standard deviation (SRSD), defined as the range of mean values divided by the average standard deviation of the voxel data for each group and modality.

Results: The average number of voxels for each bin was 578 in Group A, 1648 in Group B, and 2198 in Group C. Proportional to the number of voxels, Group C demonstrated the strongest linearity (R² = 0.82, 0.81, and 0.85 for CHU, ED, and PD) followed by Group B (R² = 0.61, 0.61, and 0.76 for CHU, ED, and PD) and Group A (R² = 0.48, 0.58, 0.46, for CHU, ED, and PD).

SRSD values for PD were variable across pooling strategies (0.35, 0.99, and 0.93 for Group A, B, and C), whereas SRSD for CHU and ED decreased with increasing pooling (1.40, 0.52, and 0.42 for CCT and 1.42, 0.99, and 0.82 for ED).

Conclusion: Our results show that CHU and calculated ED and PD values correlate strongly with temperature. Differences in R² across groups indicate that tissue heterogeneity presents a significant challenge for CT thermometry; however, voxel pooling substantially improves performance. ED and PD exhibited lower noise than CHU. More generalizable models and phantoms are necessary to establish the feasibility of using spectral CT for thermometry.