High-Dimensionality Imaging: What More Does It Give Us?
Ricardo Otazo1
1Memorial Sloan Kettering Cancer Center, United States

Synopsis

A recent paradigm shift in MRI has seen the capture of multiple dynamic dimensions in a single acquisition. At first glance, adding new dimensions would appear to make MRI more challenging, but recent developments in compressed sensing and low-rank tensor imaging have shown that this multidimensional image structure can be exploited to improve over conventional imaging performance and enable access to new physiological information of clinical interest. This talk will present the most significant developments in this paradigm shift along with relevant clinical applications.

Target audience

Researchers and clinicians who wish to learn about the advantages and new developments in high-dimensional MRI

Objectives

  1. Learn the clinical added value of high-dimensional imaging
  2. Understand the limitations of conventional dynamic MRI
  3. Learn the advantages of using compressed sensing, low-rank matrix and tensor imaging, and deep learning to overcome the curse of dimensionality and enable high-dimensional MRI with reasonable scan times and simultaneous high spatial and temporal resolution

High-dimensional MRI: New developments and new applications

Given that the human body is a dynamic system, high-dimensional imaging is desirable to capture multiple overlapped dynamics as a single multidimensional entity. For example, the cardiovascular system involves cardiac motion, respiratory motion and blood flow. Moreover, MRI offers flexible imaging contrast and quantification of tissue parameters requires the acquisition of additional dimensions. High-dimensional imaging including physiological motion, blood flow and tissue parameters can lead to comprehensive examinations and thus to overall better understanding of several diseases. A major obstacle for conventional high-dimensional imaging is the well- known “curse of dimensionality”, this is, the amount of data required increases exponentially as the imaging dimensionality increases.
Recent developments aim to exploit the structure or correlations in the high-dimensional data to reduce sampling requirements. For example, in the case of motion, the same anatomy is moving, which results in extensive temporal correlations between different motion states, and performing a 4D reconstruction (3D+time) that explicitly exploits sparsity along the fourth dimension can be more efficient that performing one 3D reconstruction that collects data only when the organ is in one pre-determined position (e.g., respiratory gating). This talk will discuss recent developments in high-dimensional imaging such as low-rank matrix imaging (1), multidimensional compressed sensing (2-4), combinations of low-rank imaging and compressed sensing (5-6), low-rank tensor imaging (7), and more recently deep learning (8). Application of the new high-dimensional MRI paradigm to clinical problems of interest will be also presented to analyze the potential added value compared to conventional dynamic MRI techniques.

Acknowledgements

No acknowledgement found.

References

  1. Liang ZP. Spatiotemporal imaging with partially separable function. Proc. IEEE ISBI 2007, pp. 988– 991.
  2. Feng L, Axel L, Chandarana H, Block KT, Sodickson DK, Otazo R. XD-GRASP: Golden-angle radial MRI with reconstruction of extra motion-state dimensions using compressed sensing. Magn Reson Med. 2016 Feb;75(2):775-88.
  3. Feng L, Coppo S, Piccini D, Yerly J, Lim RP, Masci PG, Stuber M, Sodickson DK, Otazo R. 5D whole-heart sparse MRI. Magn Reson Med. 2018 Feb;79(2):826-838.
  4. Cheng JY, Zhang T, Alley MT, Uecker M, Lustig M, Pauly JM, Vasanawala SS. Comprehensive Multi-Dimensional MRI for the Simultaneous Assessment of Cardiopulmonary Anatomy and Physiology. Sci Rep. 2017 Jul 13;7(1):5330.
  5. Lingala S, Hu Y, Dibella E, Jacob M. Accelerated dynamic MRI exploiting sparsity and low-rank structure: - SLR. IEEE Trans Med Imag 2011;30(5):1042-1054.
  6. Otazo R, Candès E, Sodickson DK. Low-rank plus sparse matrix decomposition for accelerated dynamic MRI with separation of background and dynamic components. Magn Reson Med. 2015 Mar;73(3):1125-36.
  7. Christodoulou AG, Shaw JL, Nguyen C, Yang Q, Xie Y, Wang N, Li D. Magnetic resonance multitasking for motion-resolved quantitative cardiovascular imaging. Nat Biomed Eng. 2018 Apr;2(4):215-226.
  8. Chen Y, Shaw JL, Xie Y, Li D, Christodoulou AG. Deep learning within a priori temporal feature spaces for large-scale dynamic MR image reconstruction: Application to 5-D cardiac MR Multitasking. Med Image Comput Comput Assist Interv. 2019 Oct;11765:495-504
Proc. Intl. Soc. Mag. Reson. Med. 28 (2020)