Characterizing geometrical accuracy in clinically optimized 7T and 3T MR images for high-precision radiation treatment of brain tumours


Characterizing geometrical accuracy in clinically optimized 7T and 3T MR images for high-precision radiation treatment of brain tumours

Peerlings, J.; Compter, I.; Janssen, F.; Wiggins, C.; Mottaghy, F.; Lambin, P.; Hoffmann, A.

Purpose
In neuro-oncology, 3 Tesla (3T) MRI is the current clinical standard for tumor localization, radiotherapy volume delineation and stereotactic (radio)surgery. With superior SNR and image resolution, anatomical 7T MRI can visualize micro-vascularization in glioblastomas potentially allowing improved target volume delineation. However, concerns regarding geometrical distortion (GD) with increasing field strength (B0) are detrimental for applications of 7T MRI in image-guided interventions. For high-precision treatment strategies, the spatial integrity of anatomical images needs to be warranted within ±1 mm. The aim of the study was to evaluate B0- and sequence-related GD in clinically relevant 7T pulse sequences and compare it to equivalent 3T sequences, and CT images.

Material & Methods
To quantify B0- and sequence-related GD in T1-GRE, T1-TFE, T2-TSE, T2-TSE FLAIR on 7T pulse sequences, a dedicated anthropomorphic head-phantom (CIRS Model 603A) was used. The phantom is composed of bone- /soft-tissue equivalent materials and contains a rigid 3D grid (3 mm rods spaced 15 mm apart). System-based distortion correction methods were applied to restore the gradient uniformity for 3T and 7T. For all CT and MR images, 436 points of interests (POIs) were defined by manual reconstruction of the 3D grid points in the respective images. GD was assessed in 3 ways. Firstly, global GD was estimated by the mean absolute difference (MADglobal) between the measured and the true Euclidian distances of all unique combinations of POIs, independent of location within the phantom. Secondly, local GD was estimated by MADlocal between the measured and the true Euclidian distances of all POIs relative to the magnetic field isocenter. Thirdly, a distortion map was created by evaluating 3D displacement vectors for each individual grid point.

Results
MADglobal in 3T and 7T images ranged from 0.19−0.75 mm and 0.27−1.91 mm, respectively, and was more pronounced than in CT images. CT was not completely free of GD with MADglobal ranging from 0.14−0.64 mm. B0-related GD was larger in 7T than in 3T MRI with MADlocal ranging from 0.11-0.73 mm and 0.21-1.81 mm, respectively (p<0.05). MADlocal increased with increasing distance from the magnetic isocenter and largest GDs were noted at the level of the skull in T1-TFE (Fig. 1). MADlocal was <1 mm for all sequences up to 68.7 mm from the isocenter. Sequence-related GD at 7T was prominent in T1-TFE and significantly differed from other 7T sequences (p<0.001). Figure 2 indicates an anisotropic distribution of GD in T1-TFE with increasing GD along the frequency-encoding direction.

Conclusion
System-related GD was present in all 3T and 7T MR images but remained within the 2 mm tolerance limit. Near the magnetic isocenter, 7T anatomical images showed no difference in geometric reliability to 3T MR images. Careful selection of 7T pulse sequences and judicious use of GD correction methods can warrant the geometrical quality required for incorporation of 7T MR into image-guided interventions.

Keywords: ultra-high field MRI; radiotherapy; brain tumours; geometrical image distortion

Permalink: https://www.hzdr.de/publications/Publ-24245
Publ.-Id: 24245