Abstract
The efficient content-based image retrieval of biomedical images is a challenging problem of growing interest in the research community. This book chapter describes a framework with two segmentation methods to analyze X-ray images of the spinal columns in order to extract vertebra regions and contours. The authors describe an application of the proposed methods which consists on an evaluation of vertebra motion induced by their movement between two or several positions. Their framework permits to extract the parameters determining vertebral mobility and its variation during flexion-extension movements. The first approach on our framework consists of a new contour vertebra detection technique using a polar signature system combined with a template matching process. This approach is based on a preliminary selection of vertebra regions. The second approach of our framework is based on automatic corner points of interest detection using the Harris corner detector.
TopIntroduction
Most of the existing research in medical image segmentation has focused on magnetic resonance (MR) and computed tomography (CT). Much less work has been done in the segmentation of X-ray images. Nevertheless, medical specialists often examine X-rays images of spinal columns to determine the presence of abnormalities or dysfunctions and to analyse the vertebral mobility. To help them to establish a good diagnosis, there exist medical image processing and analysis applications to automate different tasks dealing with the interpretation of these images. In our work, we use X-ray image instead of other kinds of medical images like CT – computed tomography – or MR – magnetic resonance – in order to avoid the high level of radiation received by the patient in the case of magnetic resonance or computed tomography. Another reason is the large quantity of information to be acquired and processed, and the cost of these methods which make them less functional.
In this work we propose a framework for vertebral mobility analysis using vertebra orientation angles applied to X-ray images of the cervical, lumbar and dorsal vertebrae. The purpose of the diagnosis is to extract some quantitative measures of particular changes between images acquired at different moments.
Extraction of vertebra contours from X-ray images is an important first step in computer analysis applied to medical images of the spinal column. Several methods have been applied to vertebra segmentation (Duncan & Ayache, 2000). Techniques using Hough Transform (Howe, Gururajan, Sari-Sarraf, & Long, 2004; Tezmol, Sari-Sarraf, Mitra, Long, & Gururajan, 2002), deformable models (Pham, C., & Prince, 2000) such as active contour (Kass, Witkin, & Terzopoulos, 1987), active shape models (Cootes, Hill, Taylor, & Haslam, 1994), and level set methods (Sethian, 1996; Lam & Yan, 2007) are some examples of the various approaches developed.
Several methods have been proposed in the literature to analyse and to extract vertebra contours from X-ray images (Rico, Benjelloun, & Libert, 2001; Benjelloun, Tellez, & Mahmoudi, 2006). Extensive research has been done by Long et al. (Long R., L., & G.R., Thoma, 2000) and (Long R., L., & G.R., Thoma, 2001) to automatically identify and classify spinal vertebrae, where they formulate the problem of spine vertebra identification in three levels of processing: in the first stage a heuristic analysis combined with an adaptive thresholding system is used to obtain basic orientation data, providing basic landmarks in the image; in the second stage, boundary data for the spine region of interest is defined by solving an optimization problem; in the third a deformable template processing is used to locate individual vertebra boundaries at finely grained level. The main drawback of this approach is the requirements in terms of greyscale thresholding. Stanley and Long (Stanley, Long, Antani, S, GR, & Edward, 2004) proposed a new method of subluxation detection. They used the spatial location of each vertebra in the spinal column and the variation in its position. They applied a second order spinal column approximation by using the vertebral centroid. The goal of their approach was to quantify the degree to which vertebra areas within the image were positioned on their posterior sides. In another work, Rodney and Thoma (Long .R., L., & G.R., Thoma, 1999) described a reliable method for automatically fixing an anatomy-based coordinate system in the image with an adaptive thresholding system.
In this work, we propose to investigate a framework with two segmentation methods. The first is based on a polar signature system combined with a template matching process. The second one is based on Harris corner detector allowing determining the corner points of interest in order to extract spine position.
Key Terms in this Chapter
Medical X-Ray Images: An X-ray is a type of ray used in medical imaging to diagnose several types of diseases such as breast cancer or vertebral anomalies. X-ray image are produced by placing an X-ray source on one side of the area of the body to be imaged and an X-ray film or detector on the other side. The body casts a shadowy image onto the detector, and the image is produced as a film or digital picture. The X-ray image is viewed by a radiologist, who interprets it.
Vertebral Mobility Analysis: Vertebral mobility analysis consists on the analysis of the variation of the orientation angles corresponding to the spinal column in the three positions: neutral, extension and flexion position.
Polynomial Fitting: Contour modelling using polynomial curve fitting is frequently used in computer vision and image analysis in order to obtain curves that best approximate a given collection of points of contour. We use the polynomial fitting, for example, to close the open contour.
Image Segmentation: The purpose of image segmentation is to partition an image into meaningful regions with respect to a particular application. The segmentation is based on measurements taken from the image and might be grey level, colour, texture, depth or motion
Template Matching: Template matching is a technique in Digital image processing for finding small parts of an image which match a template function or image. In our work, the goal of the proposed template matching process is to find the positions on the image which are best correlated with the template function.
Corner Detection: Corner detection is an approach used within computer vision systems to extract the points of interests on an image. These points are invariant under translation and rotation. Corner detection is frequently used in motion detection, image matching, and object recognition.
Edge Detection: is a terminology in image processing and computer vision which consists on the location of edges by identifying points in a digital image at which the image brightness changes sharply or more formally has discontinuities. The approaches based on first or second derivative of grey-scale intensity in the neighbourhood of each pixel are usually used.