Determining Spatial Resolution from Brain Imaging for Automated Quality Assurance of Magnetic Resonance Imaging
Abstract
Purpose
Magnetic resonance imaging quality assurance (MRI QA) currently relies on phantoms for characterizing image quality metrics. The growing prevalence of MRI-guided radiotherapy will increase the need for efficient and robust MRI QA. Using patient imaging for QA instead of phantom scans could reduce the physics workload and additionally enable continuous performance monitoring. Aiming to develop in vivo QA techniques, the objective of this study was to estimate spatial resolution, a fundamental imaging parameter measured daily/weekly for MRI QA, from brain imaging by characterizing edge sharpness.
Methods
Images from 30 brain tumour patients were retrospectively curated from a 1.5T MRI-linac (N=15) and MR-simulators of field strengths 0.5T (N=5), 1.5T (N=5), and 3T (N=5). Protocols included T1-weighted, diffusion-weighted, and multi-echo spin echo images with spatial resolutions between 1.0-3.0 mm (50 images total). An analytic edge-spread function was fitted to the intensity profile perpendicular to the anatomical boundary between the corpus callosum and lateral ventricles. The fitting parameter corresponding to edge sharpness was taken as the in vivo spatial resolution estimate. The correspondence between the estimated and nominal acquisition spatial resolution (average in-plane voxel dimension) was compared by Bland-Altman analysis and the Pearson correlation coefficient.
Results
The agreement between estimated and nominal spatial resolutions was statistically significant (Pearson r=0.97, p<.001). The mean bias was 0.052 mm (below 5%), and the limits of agreement were [-0.26 mm, 0.36 mm] (<36% maximum bias). There was a trend towards proportional bias, with more positive bias for smaller voxel sizes, though this did not reach statistical significance (p=.071).
Conclusion
Spatial resolution of MRI can be measured from patient imaging. In vivo QA could enable continuous performance monitoring in busy MRI-guided radiotherapy departments. Future work will include developing in vivo QA techniques for other image quality metrics including signal-to-noise ratio, geometric distortion, and low-contrast object detectability.