Characterizing Dynamic MRI Using Objective Image Quality Metrics
Quantitative mapping in dynamic magnetic resonance imaging (MRI) aims to recover quantitative parameters describing properties of patient tissues. However, quantitative mapping in dynamic MRI is difficult due to limitations imposed on temporal resolution to preserve image quality. Compressed sensing (CS) is a technique allow- ing increased temporal resolution while maintaining higher image quality than was traditionally possible. However, the compromise between quantitative mapping performance and image quality in dynamic CS-MRI remains to be explored. In this thesis we present three studies, building up towards the application of objective image quality metrics (IQMs) to characterize the performance of quantitative mapping in dynamic CS-MRI. We chose five IQMs for study: the root mean squared error (RMSE), the gradient magnitude similarity deviation (GMSD), the structural similarity index (SSIM), the multi-scale SSIM, and the information-weighted SSIM. We first explored the design of a k-space acquisition scheme designed specifically for dynamic CS-MRI, utilizing the IQMs to assess the performance of each design. It was observed that designs did not differ significantly in performance within a fixed k-space undersampling factor. Incorporating knowledge from the k-space acquisition scheme design study, we next investigated the properties of each IQM in static CS-MRI of the pelvis. Using simulated k-space acquisitions, the properties of the IQMs in static CS-MRI of 15 pelvic image data sets were investigated. It was found that the GMSD had the largest inter-image variation, followed by the RMSE and then the SSIM family of IQMs. Furthermore, the ability of each IQM to select favourable CS reconstruction parameters under different reconstruction conditions as undersampling of k-space increased was observed and characterized. With knowledge of the behaviour of the IQMs in static CS-MRI, a first-generation simulation framework for quantitative mapping in dynamic CS-MRI was developed. We present preliminary results to validate the performance of the simulation framework. The ability to investigate changing CS reconstruction parameter needs for both an emphasis on image quality and an emphasis on quantitative mapping performance in dynamic CS-MRI was demonstrated. Finally, evidence of correlations between IQM scores and quantitative mapping performance in dynamic MRI applications was demonstrated.