Second-Order Regression-Based MR Image Upsampling

Abstract

The spatial resolution of magnetic resonance imaging (MRI) is often limited due to several reasons, including a short data acquisition time. Several advanced interpolation-based image upsampling algorithms have been developed to increase the resolution of MR images. These methods estimate the voxel intensity in a high-resolution (HR) image by a weighted combination of voxels in the original low-resolution (LR) MR image. As these methods fall into the zero-order point estimation framework, they only include a local constant approximation of the image voxel and hence cannot fully represent the underlying image structure(s). To this end, we extend the existing zero-order point estimation to higher orders of regression, allowing us to approximate a mapping function between local LR-HR image patches by a polynomial function. Extensive experiments on open-access MR image datasets and actual clinical MR images demonstrate that our algorithm can maintain sharp edges and preserve fine details, while the current state-of-the-art algorithms remain prone to some visual artifacts such as blurring and staircasing artifacts.

Noise-Robust MRI Upsampling Using Adaptive Local Steering Kernel

“…On the other hand, coefficient computation is also a fundamental issue for interpolation-based MR image upsampling. Previous studies [8], [10] have advocated that using high-order patch statistics in deriving interpolation weights can be a big help for reconstructing high frequency details. In other words, weights should be adapted to high-order patch features so that details are preserved in the context of interpolation.…”

“…To solve these two problems, several researchers have proposed methods that performed adaptive interpolation [3]- [10]. In the pioneering work of Manjón et al [3], sampled voxels were selected from a large cube centered on the target voxel.…”

“…For interpolation methods, they cannot consistently reconstruct high-frequency details from low-resolution (LR) images. This is because interpolation-based algorithms all belong to the framework of zero-order regression estimation [10]. To this end, a regression-inspired upsampling method using second-order polynomials was proposed in our previous study [10].…”

“…This is because interpolation-based algorithms all belong to the framework of zero-order regression estimation [10]. To this end, a regression-inspired upsampling method using second-order polynomials was proposed in our previous study [10]. However, both adaptive interpolation methods and our previous work fail to account for the noise present in MR images, mainly produced by echo planar imaging [11].…”

“…Image denoising is typically used in these methods to remove noise beforehand. It has been demonstrated that a simultaneous interpolation and denoising eventually outperforms an asynchronous framework [10]. Motivated by this, we propose to develop a new unified MR image upsampling framework that simultaneously performs HR image reconstruction, denoising, and high-frequency detail enhancement, and thereby leads to more accurate treatment contours.…”