Fast matrix-free methods for model-based personalized synthetic MR imaging

TL;DR

This paper introduces fast, matrix-free estimation methods for model-based synthetic MR imaging that outperform existing approaches in clinical settings, enabling real-time image synthesis with reliable uncertainty estimates.

Contribution

The paper develops theoretically sound, computationally efficient matrix-free algorithms for model-based synthetic MR imaging, improving speed and accuracy over prior methods.

Findings

Superior performance in clinical settings compared to existing methods

Fast synthesis during patient visits with high accuracy

Provides reliable standard error estimates for regional contrasts

Abstract

Synthetic Magnetic Resonance (MR) imaging predicts images at new design parameter settings from a few observed MR scans. Model-based methods, that use both the physical and statistical properties underlying the MR signal and its acquisition, can predict images at any setting from as few as three scans, allowing it to be used in individualized patient- and anatomy-specific contexts. However, the estimation problem in model-based synthetic MR imaging is ill-posed and so regularization, in the form of correlated Gaussian Markov Random Fields, is imposed on the voxel-wise spin-lattice relaxation time, spin-spin relaxation time and the proton density underlying the MR image. We develop theoretically sound but computationally practical matrix-free estimation methods for synthetic MR imaging. Our evaluations demonstrate superior performance of our methods in currently-used clinical settings when compared to existing model-based and deep learning methods. Moreover, unlike deep learning approaches, our fast methodology can synthesize needed images during patient visits, with good estimation and prediction accuracy and consistency. An added strength of our model-based approach, also developed and illustrated here, is the accurate estimation of standard errors of regional contrasts in the synthesized images. A R package $symr$ implements our methodology.