Fast Automatic Vertebrae Detection and Localization in Pathological CT Scans - A Deep Learning Approach
Automatic detection and localization of vertebrae in medical images are highly sought after techniques for computer-aided diagnosis systems of the spine. However, the presence of spine pathologies and surgical implants, and limited field-of-view of the sp
- PDF / 557,508 Bytes
- 9 Pages / 439.363 x 666.131 pts Page_size
- 75 Downloads / 326 Views
1 Department of Electrical and Computer Engineering Department of Mechanical Engineering, University of British Columbia, Vancouver, Canada
Abstract. Automatic detection and localization of vertebrae in medical images are highly sought after techniques for computer-aided diagnosis systems of the spine. However, the presence of spine pathologies and surgical implants, and limited field-of-view of the spine anatomy in these images, make the development of these techniques challenging. This paper presents an automatic method for detection and localization of vertebrae in volumetric computed tomography (CT) scans. The method makes no assumptions about which section of the vertebral column is visible in the image. An efficient approach based on deep feed-forward neural networks is used to predict the location of each vertebra using its contextual information in the image. The method is evaluated on a public data set of 224 arbitrary-field-of-view CT scans of pathological cases and compared to two state-of-the-art methods. Our method can perform vertebrae detection at a rate of 96% with an overall run time of less than 3 seconds. Its fast and comparably accurate detection makes it appealing for clinical diagnosis and therapy applications. Keywords: Vertebrae localization, vertebrae detection, deep neural networks.
1
Introduction
Automatic vertebrae detection and localization in spinal imaging is a crucial component for image-guided diagnosis, surgical planning, and follow-up assessment of spine disorders such as disc/vertebra degeneration, vertebral fractures, scoliosis, and spinal stenosis. It can also be used for automatic mining of archived clinical data (PACS systems in particular). Furthermore, it can be a pre-processing step for approaches in spine segmentation, multi-modal registration, and statistical shape analysis. The challenges associated with building an automated system for robust detection and localization of vertebrae in the spine images arise from: 1) restrictions in field-of-view; 2) repetitive nature of the spinal column; 3) high inter-subject variability in spine curvature and shape due to spine disorders and pathologies; and 4) image artifacts caused by metal implants. c Springer International Publishing Switzerland 2015 N. Navab et al. (Eds.): MICCAI 2015, Part III, LNCS 9351, pp. 678–686, 2015. DOI: 10.1007/978-3-319-24574-4_81
Fast Automatic Vertebrae Detection and Localization
679
Several methods have been proposed in the literature for automatic vertebrae detection and localization in Computed Tomography (CT) [8,9,11,16,10] and Magnetic Resonance Imaging (MRI) volumes [10,15,14,1]. Several studies either concentrate on a specific region of the spine, or make assumptions about the visible part of the vertebral column in the image. A few recent studies claim handling arbitrary-field-of-view scans in a fully-automatic system [8,9,11,16]. The methods proposed in [8] and [16] rely on a generative model of shape and/or appearance of vertebrae. As a result, these methods may be challenged with patho
Data Loading...
