View of the Micro-CT Data, Where the Blue and Green Marked the Spongy Structure Caused by the Cancellous Bone

View of the Micro-CT Data, Where the Blue and Green Marked the Spongy Structure Caused by the Cancellous Bone

Liu et al. EURASIP Journal on Image and Video EURASIP Journal on Image Processing (2019) 2019:60 https://doi.org/10.1186/s13640-019-0456-1 and Video Processing RESEARCH Open Access Mice endplate segmentation from micro-CT data through graph-based trabecula recognition Shi-Jian Liu1†, Zheng Zou2†, Jeng-Shyang Pan1,3,4* and Sheng-Hui Liao5 Abstract Though segmentation of spinal column from medical images have been intensively studied for decade, most of the works were concentrated on the segmentation of the vertebral body and arch, instead of the endplate. Recently, the increasing study on degeneration analysis of vertebra and intervertebral disc (IVD) make endplate segmentation as important as others. While the accurately segmentation of mice endplate from micro computed tomography (CT) images is challenging. The major difficulties include potential high system payload and poor run-time efficiency resulting from high-resolution micro-CT data, highly complicated and variable shape of the vertebra tissues, and the ambiguous segmentation boundary due to the similarity of spongy structures inside both the endplate and its adjacent vertebral body. To solve the problems, the core idea of the proposed method is to identify trabeculae between the endplate and the body through a graph-based strategy. In addition, in order to reduce the data complexity, an endplate-targeted region of interest (ROI) extraction method is introduced according to the analysis of spatial relationship and variety of bone density of vertebra. Furthermore, shape priori of endplate in both two-dimensional and three-dimensional are extracted to assist in the segmentation. Finally, an iterative cutting procedure is implemented to produce the final result. Experiments were carried out which validate the performances of the proposed method in terms of effectiveness and accuracy. Keywords: Vertebral endplate, Trabecular bone, Segmentation, Harmonic field, Graph cuts, Micro-computed tomography 1 Introduction disc, distributing applied loads evenly to the underlying Demanded by image-based spine assessment, biomechan- vertebra, but also serves as a semi-permeable interface ical modeling and surgery simulation, segmentation of that allows the transfer of water and solutes, preventing spinal column from medical images have been intensively the loss of large proteoglycan molecules from the disc studied for decades. In former researches, most of the [1, 2]. Recently, it is also believed that there has a close works were concentrated on segmenting vertebral tis- relationship between the harm of endplate and osteoporo- sues such as the body and arch, while vertebral endplate sis [3]. Therefore, endplate segmentation could become segmentation were rarely mentioned. However, endplate an important pre-requisite of vertebra/IVD degeneration is as important as the others. As a transitional zone analysis and many other applications, which is a major between the intervertebral disc (IVD) and vertebra, it not motivation of this work. only plays an important role in containing the adjacent Up to now, endplate segmentation from computed tomography (CT) images accurately is still a challenging *Correspondence: [email protected] task in practice, which relies heavily on knowledge, expe- †Shi-Jian Liu and Zheng Zou contributed equally to this work. rience, and manual works [4]. The main difficulties can be 3College of Computer Science and Engineering, Shandong University of Science and Technology, 266590 Qindao, China concluded as follows. 4 Department of Information Management, Chaoyang University of • Technology, Taizhong, Taiwan Potential high system payload and poor run-time Full list of author information is available at the end of the article efficiency. The average thickness of endplate is © The Author(s). 2019 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. Liu et al. EURASIP Journal on Image and Video Processing (2019) 2019:60 Page 2 of 13 generally less than 80 μm; therefore, a high- construct a three-dimensional (3D) surface by propagat- resolution imagery such as the micro-CT technique ing the ray to detect the accurate gap area between two is required in order to achieve a high-accuracy processes. segmentation. While, high resolution often leads to A common strategy to reduce the data complexity is high system payload and poor run-time efficiency. by limiting the algorithm applied to a relatively small • Highly complicated and variable shape of the region of interest (ROI) [13]. In order to identify the vertebra tissues. The shape of vertebra including the ROI, Athertya et al. [5] extract the Harris corners [14] endplate, body and arch are highly complicated and among the possible vertebra region to detect the vertebra variable. For example, the mice vertebra mice we test range. However, the detection needs several interactions have transverse processes relatively longer and of and training in advance, meanwhile the accuracy of detec- larger curvature comparing to the human’s. tion depends on the selection of multi-stage seed points. Therefore, mice vertebral body and arch are more Cheng et al. [6] detect the vertebra region by a proba- likely to be interrupted with each other; it is more bilistic map computed from a voxel-wise classifier and use difficult to separate them properly. Besides, existing mean shift algorithm to estimate the ROI after annota- algorithms designed for human vertebrae may not be tions. suitable for our case anymore. In our work, we also need to isolate individual vertebra • Ambiguous segmentation boundary. Both the from a given vertebral column, to extract the ROI for data endplate and body are composed of spongy complexity reduction, while more importantly, to segment cancellous bone (see blue and green region marked in the endplate, and we made it in a endplate-targeted way. Fig. 1a), which makes the segmentation boundary Specifically, taking micro-CT images as input, we aim to ambiguous. Though there exists a gap area (labeled develop an intelligent framework to accurately segment with a yellow dotted line in Fig. 1b) between them, theendplatefromtheothers.Andthecoreideaisto however, the gap is considerably narrow and divided recognize the trabeculae (e.g., marked by red circles in by trabeculae, which means it is difficult to find a Fig. 1b) within the narrow gap formed by the endplate and proper continuous curve as a segmentation boundary. its adjacent vertebral body by a graph-based algorithm. The proposed framework can be separated into four parts, As we mentioned above, though few works could be which are (1) pre-processing, (2) priori shape extraction, found on endplate segmentation, plenty of researches (3) gap trabecula detection, and (4) trabecula cutting have been done to achieve detection and segmentation of and refining. Particularly, we firstly isolate each vertebra vertebrae [5–9], spinal canals [10], and spinal cord [11]. and identify the ROI for endplate in a top-down basis For the purpose of separating the neighboring vertebrae (Section 2.1). Secondly, shape priori in both 2D and 3D are roughly, Zhao et al. [12] proposed a neighboring point rat- introduced to offer constraints for later graph cuts based ing method based on constructing membership grade to trabeculae recognition (Section 2.2). After that, a graph form a virtual plane to cut adjacent processes. The method will be constructed based on mask skeletonization before proposed by Kim et al. [9]searchesarayemittedfromthe the graph cuts based trabeculae recognition (Section 2.3). stared point among the center axis of the spine and further Finally, we find an optimal cutting strategy to remove the Fig. 1 Demonstration of endplate segmentation difficulties. a A sagittal view of the micro-CT data, where the blue and green marked the spongy structure caused by the cancellous bone. b Zoom in view of the endplate and its adjacent vertebral body region, where two of the gap trabecular and the gap line are labeled in red and yellow respectively Liu et al. EURASIP Journal on Image and Video Processing (2019) 2019:60 Page 3 of 13 redundant connections properly (Section 2.4). An accu- from both the top and bottom part to the center part rate result can be achieved after these processes, which in axial planes slice by slice as indicated by blue arrows is ready for subsequent operations such as the 3D recon- in Fig. 4. Because in this way, we could find two regions struction. indicated by R1 and R2 in Fig. 4 which cover the entire endplate. Then, based on the projection of bone regions 2 Material and methods within R1 and R2, we can easily find a circle (e.g., C1 in The experimental data we used were collected from Fig. 4) which can be used to define the target cylinder. the animal laboratory of spinal surgery of the Xiangya The last step produces the finest ROI by further con- Hospital. Micro-CT technique is adopted due to the high- firming the proper range of the axial slices. It travels each quality imagery requirement. To capture the data, mice slice in the way as described in the second step, while of different ages were fixed in a slot scanned by a Bruker other than analyzing the topology, bone density ρ(Ii) of SkyScan1172 micro-CT scanner. The in-plane resolution each slice i will be calculated according to Eq. 1,and of the image is 7.27μm × 7.27 μm, and the slice spacing then the variation of bone density will be used to find the is 7.27 μm. Figure 2 demonstrates a typical data of the rightful indexes. inputs, which contains 4 lumbar vertebrae. NF (Ii) 2.1 Pre-processing ρ(I ) = × 100% (1) i A (I ) The aim of the pre-processing is to get proper ROI for F i the endplates.

View Full Text

Details

  • File Type
    pdf
  • Upload Time
    -
  • Content Languages
    English
  • Upload User
    Anonymous/Not logged-in
  • File Pages
    13 Page
  • File Size
    -

Download

Channel Download Status
Express Download Enable

Copyright

We respect the copyrights and intellectual property rights of all users. All uploaded documents are either original works of the uploader or authorized works of the rightful owners.

  • Not to be reproduced or distributed without explicit permission.
  • Not used for commercial purposes outside of approved use cases.
  • Not used to infringe on the rights of the original creators.
  • If you believe any content infringes your copyright, please contact us immediately.

Support

For help with questions, suggestions, or problems, please contact us