Enhancement of Perivascular Spaces in 7 T MR Image using Haar Transform of Non-local Cubes and Block-matching Filtering

Scientific Reports, Aug 2017

Perivascular spaces (PVSs) in brain have a close relationship with typical neurological diseases. The quantitative studies of PVSs are meaningful but usually difficult, due to their thin and weak signals and also background noise in the 7 T brain magnetic resonance images (MRI). To clearly distinguish the PVSs in the 7 T MRI, we propose a novel PVS enhancement method based on the Haar transform of non-local cubes. Specifically, we extract a certain number of cubes from a small neighbor to form a cube group, and then perform Haar transform on each cube group. The Haar transform coefficients are processed using a nonlinear function to amplify the weak signals relevant to the PVSs and to suppress the noise. The enhanced image is reconstructed using the inverse Haar transform of the processed coefficients. Finally, we perform a block-matching 4D filtering on the enhanced image to further remove any remaining noise, and thus obtain an enhanced and denoised 7 T MRI for PVS segmentation. We apply two existing methods to complete PVS segmentation, i.e., (1) vesselness-thresholding and (2) random forest classification. The experimental results show that the PVS segmentation performances can be significantly improved by using the enhanced and denoised 7 T MRI.

A PDF file should load here. If you do not see its contents the file may be temporarily unavailable at the journal website or you do not have a PDF plug-in installed and enabled in your browser.

Alternatively, you can download the file locally and open with any standalone PDF reader:

https://www.nature.com/articles/s41598-017-09336-5.pdf

Enhancement of Perivascular Spaces in 7 T MR Image using Haar Transform of Non-local Cubes and Block-matching Filtering

Abstract Perivascular spaces (PVSs) in brain have a close relationship with typical neurological diseases. The quantitative studies of PVSs are meaningful but usually difficult, due to their thin and weak signals and also background noise in the 7 T brain magnetic resonance images (MRI). To clearly distinguish the PVSs in the 7 T MRI, we propose a novel PVS enhancement method based on the Haar transform of non-local cubes. Specifically, we extract a certain number of cubes from a small neighbor to form a cube group, and then perform Haar transform on each cube group. The Haar transform coefficients are processed using a nonlinear function to amplify the weak signals relevant to the PVSs and to suppress the noise. The enhanced image is reconstructed using the inverse Haar transform of the processed coefficients. Finally, we perform a block-matching 4D filtering on the enhanced image to further remove any remaining noise, and thus obtain an enhanced and denoised 7 T MRI for PVS segmentation. We apply two existing methods to complete PVS segmentation, i.e., (1) vesselness-thresholding and (2) random forest classification. The experimental results show that the PVS segmentation performances can be significantly improved by using the enhanced and denoised 7 T MRI. Introduction The perivascular space (PVS) is the normal anatomical structure filled with cerebrospinal fluid (CSF) in the nerve system. It was initially identified by a German pathologist, R. Virchow, and a French biologist, C.P. Robin, more than a century ago. The PVS, which is also known as the perivascular lymphatic space, is related to brain physiological and immune functions1,2,3,4. For example, the diameter of PVSs is typically less than 2mm in all age groups of healthy people. The abnormal enlargement of the diameter or the increase of the PVS number is related to aging5,6,7, cognitive degeneration8, and vessel diseases6, 9, 10. Although PVS has drawn more interest in recent years, the traditional 1.5 T and 3 T magnetic resonance (MR) scanners can hardly capture the thin structures of PVSs precisely. The new-generation 7 T MR scanner, on the contrary, owns much better capability of rendering PVSs10. However, it is still difficult to clearly distinguish the thin PVSs from the noisy background in the 7 T MR images. Figure 1(a) shows the examples of PVSs in a 7 T MR image, where those relatively bright voxels within the red dotted contours are the PVSs. We can see that the PVSs are the thin structures with weak signals in the 7 T MR image. Figure 1 (a) Illustration of PVSs in a 7 T MR image. Those relatively bright voxels within the red dotted contours are the PVSs. (b) Enhanced image by using spatial correlation filtering; and (c) enhanced image by using our proposed method. (d–f) Show the denoised images of (a–c), respectively. For these three cases, the BM4D method uses the same parameters to denoise the images. Full size image There are several reports that focus on the identification or segmentation of PVSs. For example, Wuerfel et al.11 developed a semi-automatic software that can adjust the intensity threshold for segmenting PVSs by optimizing a predefined PVS model through Markov Chain Monte Carlo. Uchiyama et al.12 used the white top-hat transform to enhance the tubular structures of PVSs, then extracted them according to intensity thresholding, and then finally identified PVSs using the geometric properties. Ramirez et al.13, 14 proposed to segment PVSs in a semi-automatic way by determining the intensity thresholds from T2-weighted and proton density (PD) images adaptively. Wang et al.15 also proposed a semi-automatic method that can adaptively adjust the intensity threshold by using gamma correction and linear mapping. Recently, Park et al.16 proposed a learning-based automatic segmentation method by training a sequence of random forest classifiers with the orientation-normalized 3D Haar features. Then, the segmentation is attained through the sequential classifiers, followed by vesselness thresholding. Although these existing methods can effectively use various characteristics of PVSs, the segmentation performance is still limited due to image quality. In particular, it is difficult to extract informative features of the thin and weak PVS structures from the noisy background. Accordingly, it is necessary to enhance image quality to effectively distinguish the PVS structures. A common image enhancement is through spatial correlation filtering, which performs convolution operation on every voxel. However, the convolution operation using a single kernel cannot adaptively amplify the PVS signals and also, at the same time, suppress the noises (see Fig. 1(b)). To effectively enhance the specific PVS-like structures, Uchiyama et al.12 incorporated a white top-hat transform method, which is relatively less sensitive to both image noise and inhomogeneous intensity condition. However, since the white top-hat transform is also (...truncated)


This is a preview of a remote PDF: https://www.nature.com/articles/s41598-017-09336-5.pdf

Yingkun Hou, Sang Hyun Park, Qian Wang, Jun Zhang, Xiaopeng Zong, Weili Lin, Dinggang Shen. Enhancement of Perivascular Spaces in 7 T MR Image using Haar Transform of Non-local Cubes and Block-matching Filtering, Scientific Reports, 2017, Issue: 7, DOI: 10.1038/s41598-017-09336-5