Coronary vessel wall MR imaging has great potential to detect coronary plaques which can be of vital importance for heart disease diagnosis. However, coronary vessel wall imaging is always time-consuming because a large amount of data is needed for clear vessel wall depiction. Compressed sensing (CS) with total variation (TV) [1] is a promising approach to accelerate MR imaging by reducing the amount of acquired data, however, TV gives equal penalty to each pixel [2] on the image, resulting in coronary vessel wall blurring. To address this issue, a new method based on CS with block-weighted total variation (BWTV) was proposed in this work. We hypothesized that the proposed method, named BWTV, applied a small weight to the small vessel wall region will help to significantly accelerate the coronary vessel wall imaging without compromising the depiction of the coronary vessel wall.

For coronary vessel wall imaging, the most important thing is to clearly visualize the tiny vessel wall which only occupies a small region on the image. Thus, a small penalization is preferred to regularize this region to preserve the edge of the vessel wall, and a large one is preferred for the other regions. The image reconstruction then can be formulated as Eq. (1).

$${u_{{\rm{rec}}}} = \mathop {\arg \min }\limits_u \left\{ {{{\left\| {{\bf{F}}u - y} \right\|}_2}^2 + \lambda {{\left\| u \right\|}_{{\rm{BWTV}}}}} \right\} (1)$$

where $$$y$$$, $$${\bf{F}}$$$ and $$$\lambda$$$ represents the sampled data in k-space, the partial Fourier encoding matrix and the regularization parameter respectively, and $$${\left\| u \right\|_{{\rm{BWTV}}}}$$$ is the BWTV norm of image $$$u$$$ and is calculated as $$${\left\| u \right\|_{{\rm{BWTV}}}} = \int {{\omega _\sigma }{{\left\| {{u_\sigma }} \right\|}_{{\rm{TV}}}}d\sigma } $$$ where $$${\left\| {{u_\sigma }} \right\|_{{\rm{TV}}}}$$$ represents the TV norm of the block $$${u_\sigma }$$$ of the image and $$${\omega _\sigma }$$$ denotes the weight of the corresponding block.

(1) Numerical simulation:

A modified Shepp-Logan image with matrix size of 256×256 was used for simulation. A tiny annulus and three bright adjoined pixels on the inner wall were placed on center of the image to simulate coronary vessel wall and plaque. Besides, two discs and one ellipse with proper gray-levels were added to reduce the image sparsity. Fourier transformed data of the simulated image was used as full k-space data. And then down-sampled data with different acceleration factors were retrospectively obtained by densely sampling 24 center phase lines and randomly down-sampling outer ones from the above data. The acceleration factors were set to 2, 4, 6 and 8.

(2) In vivo experiment:

IRB-approved 4 healthy volunteers were recruited for assessing the proposed method. A 3D dark-blood sequence with stack-of-stars sampling trajectories and a partition-first tiny-golden-angle reordering was customized for data acquisition [3,4]. The study was performed on a 3T (MAGNETOM Trio, Siemens) with a standard 12-channel coil. Scan parameters included: 3D transversal imaging to coronary vessel, TR/TE= 3.8/1.9 ms, spatial resolution=0.83×0.83×2.0 mm³, bandwidth=992 Hz/Pixel, partition number=12.radial views of a partition=240. After data acquisition, the data was retrospectively down-sampled according to acceleration factors of 2, 3 and 4, respectively.

(3) Image reconstruction:

In both simulation and in vivo experiment, two blocks were defined for image reconstruction using the proposed BWTV method. The size of the central block containing the vessel wall was empirically chosen as 81×81 in vivo data while 41×41 in simulation to make sure that the vessel wall was included. For comparison, TV and edge-preserving TV [2] were also conducted to reconstruct all data. The regularization parameter $$$\lambda $$$ was respectively tuned to achieve the best reconstruction by different methods.

Numerical simulation results were shown in Fig. 1&2. When a low acceleration factor of 2 was applied, the image was accurately reconstructed by all three methods. However, as the acceleration factor increased, the images reconstructed by TV and EPTV were quickly deteriorated, while both the coronary vessel wall and the plaque can be well reconstructed by BWTV even the acceleration factor is 8 (Fig. 1). Representative in vivo experiment results from a subject were shown in Fig. 3. The vessel wall was well reconstructed by BWTV with acceleration factor of 4 (i.e, only 60 radial views per partition), while the wall reconstructed by TV and EPTV had become heavily blurry with acceleration factor of 3. A novel method was proposed to accelerate coronary vessel wall imaging. Both simulation and in vivo experiment results demonstrate that the proposed BWTV method can decrease the amount of data for image reconstruction without compromising the visualization of coronary vessel wall. Further evaluation is required to establish its clinical value, which is our goal in next step.