Magnetic resonance imaging (MRI) is widely used to diagnostically investigate the human body. However, unlike other medical imaging modalities (e.g., X-ray CT), MRI data is not explicitly quantitative - pixel values do not directly represent physiological information. There is strong clinical interest in transforming MRI into a quantitative modality, and to measure quantities such as flow, diffusion, perfusion, T1 relaxation times, and fat/water concentrations. For these, series of images are collected (e.g., at different acquisition settings), physics-based models are fit to data, and relevant physiological parameters are extracted. During this process, however, MRI data typically passes through a complex processing pipeline. Although errors in raw MRI data are well-understood, how they propagate through this pipeline is not, and current results on these applications are typically biased and error-prone. We have developed a novel, generically applicable mathematical framework that instead directly estimates physiological parameters from the raw data, even if this data is incomplete (i.e., “undersampled”), enabling statistically optimal error handling and thus increased accuracy. We believe this statistical framework, and the novel computational method developed for practically applying it, is a key advance towards making MRI a truly quantitative diagnostic tool.
Model-Based Reconstruction Strategy for Accelerated Quantitative Magnetic Resonance ImagingTechnology #2013-081
Questions about this technology? Ask a Technology Manager