Vertebra localization and segmentation using shape based analysis and unsupervised clustering from x-ray images

Abstract

During the past few years, medical imaging has become one of the most useful tool for diagnosis of di erent diseases. These revolutions have also helped to obtain information useful in many clinical applications and diagnoses of disorders such as osteoporosis, spinal ruptures and cervical spine trauma. The cervical injuries may e ect arms, legs, and middle parts of the body. The poor contrast and noisy set of image data makes it di cult accurate vertebral detection in radiograph is a di cult task mainly due to low contrast and noisy set of image data. For the diagnosis of spinal disorders such as cervical spine trauma and whiplash, the detection and segmentation of vertebra are the fundamental tasks. The rst step in the detection process is the vertebra localization followed by segmentation. In this framework, an analysis of x-ray image is required which can be achieved only if an accu- rate localization and segmentation of cervical vertebrae is performed. Vertebrae localization and segmentation has been in research since years but the traditional techniques are either semi-automated or lack in accuracy when applied to x-ray images. To address these issues, a decision support system (DSS) is proposed for the localization and segmentation of cervi- cal vertebrae (C3 ð€€€ C7) using x-ray images. The proposed method consists of two modules vertebra localization and segmentation. The localization module of proposed system uses generalized hough transform alongwith a mean model of vertebra shape to generate hough space. In this semi-automated method, candidate voted points are obtained within a region speci ed with the help of manual mask. These points are then clustered into 5 clusters using FCM to obtain centroids of targeted ve vertebras (C3ð€€€C7). The segmentation mod- ule of proposed system uses these vertebra centroids and intervertebral points are obtained and A ne transformation is applied on these intervertebral points and centroids for the separation of vertebra regions. Experimental results show the e ciency of the proposed ap- proach. The proposed method secured localization accuracy of 93.76% when tested on 150 x-ray images of publically available database `NHANES II'.