Please use this identifier to cite or link to this item:
|Title:||Digital Image Analysis of Vertebral Bodies from Computed Radiography Images|
|Department:||Electrical and Computer Engineering|
|Abstract:||Vertebral body deformities resulting from osteoporosis are commonly observed in lateral spine radiographs. Clinically, these deformities are judged rather subjectively, both in their classification and in the assessment of the extent of deformity. In order to objectify clinical observation, morphometric measurements are a means for classifying and quantifying the presence and extent of vertebral body deformities. Drawing on the morphometric systems of Minne et al. (1988) and Smith-Bindman et al. (1991), we propose a total vertebral deformity index (TVDI) which provides a single, clinically meaningful number indicative of the extent of deformity in a vertebral column. Each of the three deformity types (wedge, biconcave, and compression) is classified and measured independently, using ratios relating measurements of anterior, mid, and posterior heights, and inferior width. Expected measurements are determined for each vertebral level, and the sum of the deviations from the measured to the expected values generates a vertebral deformity index. This proposed index addresses issues in vertebral body morphometry, including vertebral level specificity, body size differences, multiple compression effects, and multiple deformity types. Inherent in the practical use of morphometric quantification are issues of accuracy and reproducibility of the measurements, .and of the time involved in making these measurements. Digital image processing algorithms are developed to attempt automated detection and measurement of the vertebral body boundary, using Computed Radiography images of the lateral spine. Three different methods yield varying results. In addition to the characteristics of lateral spine images (such as high intensity vertebral ridges) which are used to advantage by the algorithms, all three methods must deal with problematic anatomical characteristics such as the presence of the high intensity ribs and ilium. The first method is a series of image enhancement and thresholding steps applied to each vertebra, in order to delineate its vertebral boundary in a hi-level image. This method is fairly effective, but suffers from its dependence on thresholding. The second method uses the cross-correlation measure to detect the vertebrae, given a starting vertebra, and then uses edge-gradient tracking to trace the vertebral boundary. Results of this method are promising. There remains a weakness in the cross-correlation detector, which fails to accurately locate vertebrae that are further away from the starting vertebra. The third method is an active contouring technique called Snakes, in which the vertebral boundary is represented by a deformable spline which seeks to minimize an energy functional consisting of curvature, continuity, and image energy terms. Present results display minimal convergence onto the vertebral boundaries. Future work should refine the customization of the energy functional to produce better results. At present, all three methods require some kind of user interaction. Further development may prove fruitful in reducing user interaction to achieve a truly automated system.|
|Appears in Collections:||Digitized Open Access Dissertations and Theses|
Files in This Item:
|tan_lianne_1995Sept_masters.pdf||48.13 MB||Adobe PDF||View/Open|
Items in MacSphere are protected by copyright, with all rights reserved, unless otherwise indicated.