- Poster presentation
- Open Access
Accelerated first-pass perfusion CMR using compressed sensing with regional spatiotemporal sparsity
© Chen et al; licensee BioMed Central Ltd. 2013
- Published: 30 January 2013
- Mean Square Error
- Compress Sense
- Average Mean Square Error
- Signal Intensity Change
- Soft Thresholding
CMR perfusion images demonstrate complex dynamic behavior resulting from signal intensity changes during the first pass of gadolinium as well as motion from imperfect breathholding and gating. Reconstruction algorithms such as kt-PCA and compressed sensing (CS) techniques such as kt-Sparsity and Low-Rankness (kt-SLR) assume that a few spatiotemporal basis functions can model this intricate behavior [1, 2]. However, these techniques are sensitive to motion, as the basis functions do not accurately describe the complete dynamics of the entire image set. We propose a novel method that utilizes regional sparsity by dividing the images into regions. With this approach, the simplified dynamics of smaller regions can be better described by a limited number of basis functions. This method was tested on in vivo images and simulated data, and the results were compared to kt-SLR , a CS method that uses global sparsity.
Images were spatially divided into square blocks (approximately 15×15 pixels). As small blocks have simpler dynamic patterns and are insensitive to dynamic changes in other regions of the image, they can be represented with fewer basis functions. Singular value decomposition was applied to the dynamic blocks to exploit the high spatiotemporal correlations within them. Iterative soft thresholding  was applied to filter low singular values, which primarily represent incoherent noise and aliasing. The de-aliased blocks were merged back into images using weighted averaging . Images underwent iterative CS reconstruction through the blocking, thresholding and merging procedures, subject to fidelity with collected k-space data. Four first-pass datasets (chosen to have prominent respiratory motion) and a simulated phantom featuring respiratory motion and time-varying signal intensity were retrospectively undersampled at an acceleration rate of 4 and reconstructed using the regional sparsity method and kt-SLR. Mean square error (MSE) was calculated for quantitative analysis.
A novel CS method using regional sparsity was less sensitive to motion than kt-SLR for CMR perfusion imaging. Future work includes developing improved regional separation methods, such as pattern recognition, and further improved motion compensation using regional motion tracking.
This study is funded by Siemens Medical Solutions, NIH R01 EB 001763 and American Heart Association Predoctoral Award 12PRE1204005
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.