Real time MR thermometry, usually based on the proton-resonance frequency shift, is a key aspect of MR-guided focused ultrasound (MRgFUS) procedures [1]. The desire to monitor the entire insonicated volume has led the field towards the development of rapid, 3D methods [2]; however, acquiring fully sampled 3D volumetric data to monitor heating is time consuming, and so fast methods must be developed in order to meet the spatial and temporal requirements for adequate monitoring of thermal therapy. The data acquisition efficiency of spiral trajectories is higher than that of Cartesian scanning. Therefore, spiral trajectories are an attractive way to improve temporal resolution while maintaining spatial resolution in MR thermometry [3-6]. Previously, we have implemented spiral trajectories for 2D and 3D thermometry and compared trajectory performance in terms of the focal spot size and position shift versus a Cartesian acquisition [5,6]. The purpose of this study was to implement the sequence on a real-time platform and demonstrate its preliminary effectiveness in a more clinically-relevant set-up using an in vivo porcine model for MRgFUS thalamotomy.

A 3D retraced spiral-in/out (RIO) thermometry sequence was implemented on the RTHawk platform (HeartVista, Inc.) to enable real-time sequence control and monitoring of a FUS insonication. RTHawk interfaced with a GE Discovery MR750T 3T scanner at the UVA Focused Ultrasound Center, where an Insightec ExAblate 650 focused ultrasound transducer was used to induce focal heating.

Three Yorkshire pigs were used for initial sequence evaluation. Non-ablative power settings were used to compare the temperatures measured by the 3D sequence to those measured by the standard 2D GRE thermometry sequence provided by Insightec. Briefly, a target within the thalamus was identified, and then insonicated at low power with monitoring performed by either the 2D or 3D method. After an appropriate cool-down period, the same spot was insonicated again with the same power settings and monitored by the alternate thermometry technique. Finally, ablative power settings were used to create a lesion at each spot, again monitored by either the 2D or 3D method. All insonications were 20 seconds long. Maximum temperature, measured by the single hottest pixel, and mean temperature, measured as the average of the 9 pixels centered at the hottest pixel in a single plane, were recorded. Temperature standard deviation of each thermometry image set was measured as the mean standard deviation over time of a group of 9 pixels located within the brain away from the focal spot. After FUS, the animals were moved into a head coil array for high quality post-ablation assessment, which included T2-, Diffusion-, and T2*-weighted imaging using standard sequences.

Sequence parameters for 3D RIO thermometry were: FA = 10°, TR/TE = 20.1/12.8 ms, readout length = 8 ms, interleaves = 24 over a FOV of 280 mm² for an in-plane resolution of 1.5 mm². 3D phase encodes = 16 with through-plane resolution of 2 mm, so that through-plane FOV was 32 mm. Total acquisition time per volume was 7.7 seconds. In one animal, a spectral-spatial pulse was used for water-only excitation in order to reduce artifacts stemming from fat, which necessitated an increase of TR to 22.7 ms and temporal update interval of 8.7 seconds. Sequence parameters for the 2D thermometry method were: TR/TE = 27.6/12.8 ms, pixel bandwidth = 44 Hz/px, matrix size 256 x 128 over a FOV of 280 mm² for a resolution of 1.1 x 2.2 mm², slice width of 3 mm, and update interval 3.5 seconds.

Real-time in vivo thermometry images are shown in Fig. 1. The hot spot is resolved in volumetric space, albeit with lower SNR with the spiral 3D sequence, due in part to reduced voxel size. In Fig. 2, post-ablation 3D T2-weighted images are shown along with an overlay of one of the 3D temperature maps showing good correspondence between the two. Scatter plots of the maximum and mean measured temperatures by each sequence is presented in Fig. 3. The maximum temperatures measured by the 3D sequence correlate well with those measured by the 2D sequence. The mean temporal standard deviations of the 2D and 3D sequences were 0.8°C and 1.3°C, respectively.The efficiency of spiral readouts supports rapid generation of 3D temperature maps. In vivo, we have successfully monitored the entire ablation with adequate spatial resolution to qualitatively compare the ablation area with temperature. We have achieved 1.3°C temperature accuracy with 8.7s temporal resolution. Future work will focus on improving these specifications through the use of surface coils and temporal acceleration methods [5].