Accelerated contrast-enhanced whole-heart coronary MRI using low-dimensional-structure self-learning and thresholding

Abstract

We sought to evaluate the efficacy of prospective random undersampling and low-dimensional-structure self-learning and thresholding reconstruction for highly accelerated contrast-enhanced whole-heart coronary MRI. A prospective random undersampling scheme was implemented using phase ordering to minimize artifacts due to gradient switching and was compared to a randomly undersampled acquisition with no profile ordering. This profile-ordering technique was then used to acquire contrast-enhanced whole-heart coronary MRI in 10 healthy subjects with 4-fold acceleration. Reconstructed images and the acquired zero-filled images were compared for depicted vessel length, vessel sharpness, and subjective image quality on a scale of 1 (poor) to 4 (excellent). In a pilot study, contrast-enhanced whole-heart coronary MRI was also acquired in four patients with suspected coronary artery disease with 3-fold acceleration. The undersampled images were reconstructed using low-dimensional-structure self-learning and thresholding, which showed significant improvement over the zero-filled images in both objective and subjective measures, with an overall score of 3.6 ± 0.5. Reconstructed images in patients were all diagnostic. Low-dimensional-structure self-learning and thresholding reconstruction allows contrast-enhanced whole-heart coronary MRI with acceleration as high as 4-fold using clinically available five-channel phased-array coil.

Copyright © 2012 Wiley Periodicals, Inc.

Figures

Figure 1

Profile ordering in 3D segmented acquisition through time for a) Conventional radial profile ordering for a fully-sampled acquisition, b) modified radial profile ordering with true undersampling (rate 4 is shown as an example) and c) simulated undersampling. First, a subset of the fully-sampled ky-kz lines is selected randomly based on a given undersampling rate, while ensuring that the central ky-kz region is included. Then, the selected ky-kz lines are reordered in a radial fashion prior to the acquisition. For simulated under-sampling, the data acquisition is continued after b.11 (white dots represent the lines already acquired) to acquire the rest of the k-space lines (gray dots in c), such that the k-space is fully sampled (within an elliptical window) in c.6. In both b and c, the acquisition of each shot starts from the closest point to the center and moves to the outer k-space.

Figure 2

a–e) Profile ordering in 3D segmented acquisition for one of the example patterns used in the prospectively undersampled acquisitions. The net acceleration rate is 4 with respect to the elliptical window, which corresponds to an acceleration rate of 5.1 for the whole k-space. Due to the mismatch between elliptical geometry as specified by different FOVs in y and z axes, and the radial profile ordering with the requirement of starting each shot from innermost k-space, some outer k-space lines are acquired earlier than the inner k-space lines corresponding to the same phase in the ky-kz plane (e.g. lower left quadrant of b, and upper right quadrant of d). This phenomenon applies to both the conventional and the modified radial profile ordering.

Figure 3

Effects of different profile ordering schemes in phantom (top) and in-vivo right coronary artery (RCA) imaging (bottom). (a, d) Fully-sampled; (b, e) images acquired with simulated prospective random undersampling of rate 2 with no profile ordering. Artifacts are apparent in these images even though they are fully sampled; c depicts a phantom image acquired with the proposed profile ordering with a simulated prospective random undersampling; f depicts a true undersampling of rate 2 with the proposed profile ordering and a random undersampling pattern, reconstructed using LOST. The proposed ordering scheme mitigates artifacts associated with random undersampling patterns, and improves visibly over no profile ordering.

Figure 4

An example axial slice (left) and reformatted axial image (right) depicting the left coronary system of a healthy subject using zerofilling (acquired) and LOST reconstruction. The left coronaries are better defined (image scores 3.3 ± 0.7 for LAD, 3.4 ± 1.0 for LCX) in the LOST reconstruction, but are blurry in the original acquired images (image scores 1.2 ± 0.4 for LAD, 1.4 ± 0.6 for LCX) due to the high rate of undersampling (RCA: right coronary artery, LAD: left anterior descending, LCX: left circumflex).

Figure 5

An example axial slice (left) and reformatted axial image (right) depicting the right coronary artery (RCA) of another subject using zerofilling (acquired) and LOST reconstruction. The LOST reconstruction allows better definitions (image score 3.7 ± 0.7 versus 1.5 ± 0.5) of the RCA in the proximal, mid and distal regions.

Figure 6

Reformatted LOST-reconstructed coronary MRI (left) showing occlusion of the right coronary artery (RCA). The x-ray angiography images (right) confirm this finding.

Figure 7

Two axial slices from the LOST-reconstructed 3D whole heart coronary MRI showing coronary stenosis in a) right coronary artery and b) left coronary artery with (c, d) corresponding invasive x-ray coronary angiography in a 47-year male patient with coronary artery disease undergoing coronary artery bypass graft.