MRI-guided cryoablation is a promising minimally invasive therapy with applications in musculoskeletal, liver, kidney and prostate cancer¹. To assure effective treatment, temperature feedback during these procedures is desirable but a non-invasive approach is still warranted. Previous studies have demonstrated measurable MR signal from frozen tissue using ultrashort TE (UTE) imaging and explored its potential for MR thermometry^2,3. As an important next step, we assessed in this work the feasibility and accuracy of 3D UTE MR thermometry to dynamically track temperatures across frozen tissue during cryoablation on a clinical MR system at 3T.Four identical cryoablation experiments were performed. An MR-compatible cryoneedle (IceRod, Galil Medical, Yokneam, Israel) was inserted into ex vivo porcine muscle specimens at room temperature on a 3T clinical MR system (MAGNETOM Trio, Siemens, Erlangen, Germany). Fiber optic sensors (T1, Neoptix, Quebec, Canada) were used for temperature reference (Figure 1). Two cycles of 10:3 min. freeze-thaw were applied. Continuous MR monitoring of ice progression was performed using a 3D radial ramp-sampled UTE sequence (TR/TE/FA = 59.5ms/70μs/15°, voxel size = 1.63x1.63x1.63mm, acq. time = 1:14min). Data of three experiments were used as reference sets. Signal intensity (SI) values were normalized to the baseline value before cooling and related to temperature. Data points for subzero temperatures were fitted by an exponential function. In a separate validation set, the obtained fit was used to generate MR-estimated temperature maps of the frozen tissue. Statistical analysis was performed to determine accuracy of the estimated temperature maps.In the reference sets, next to the known T1-related signal increase during cooling from room temperature to 0°C⁴, normalized SI decreased mono-exponentially with temperature for subzero conditions with the signal decay fitted by normalized SI = 1.26e^0.05T (R²=0.93) (Figure 2). Using the fit as a calibration curve, we could obtain MR-estimated temperature maps of the frozen tissue in 3D at each imaging time point in the validation set (Figure 3). MR-estimated temperatures strongly correlated with sensor readings at matched time points over the course of the cryoablation experiment (r=0.977, p<0.001) (Figure 4a). Bland-Altman analysis demonstrated good agreement between the two measures (Figure 4b). Mean difference between MR-estimated and sensor measured temperatures was –1.7±2.8°C with upper and lower limits of agreement of –7.1 and 3.8°C respectively.In this work we demonstrated the feasibility of 3D UTE MR thermometry to dynamically track temperatures in frozen tissue during cryoablation. With currently no other method existing to non-invasively measure temperatures within frozen tissue, the accuracy we found in this work could already be of value in a clinical context. From a pragmatic approach, a relative signal level as compared to baseline before cooling may be identified for which there is a high degree of certainty that temperatures are below a certain threshold, e.g. -40°C. Nevertheless, to apply this method clinically it should be validated how ex vivo calibration curves translate to the in vivo setting. Notably, in the signal peak observed with temperature decreasing from room temperature to the freezing point we observed an offset of approximately +5°C. Theoretically, maximum signal would be expected around 0°C, directly before the freezing-related signal loss. This discrepancy probably results from a mismatch between the positions of the temperature probes and the voxels used for the SI measurements. Another contributing factor that inherently affected our measurements is temperature averaging, both within a voxel due to the spatial temperature gradient as well as over time due to temperature changes occurring during image acquisition. Despite this though, a temporal resolution of 1:14min seemed to provide adequate accuracy in tracking the temperature-induced MR signal changes, even for the faster variations occurring at the transition from the freeze to thawing phase.3D MR thermometry of frozen tissue using UTE signal intensity was feasible within the time frame of a typical cryoablation procedure on a clinical MR system at 3T. Down to temperatures as low as -40°C, accuracy of the MR-estimated temperature maps was within clinically acceptable limits.