Quantitative analysis of vertebral parameters in idiopathic scoliosis using image processing techniques

Abstract

Scoliosis is a three dimensional deformity of human spine comprising coronal plane curvature and axial rotation. Imaging modalities such as radiography, computed tomography (CT) and magnetic resonance imaging (MRI) play pivotal roles in the diagnosis, monitoring and management of the scoliosis with radiography having the primary role with MRI/CT. Severity of the spine curvature in the lateral view is quantified by Cobb angle. Quantification of Cobb angle is based on the identification of vertebral endplates as well as manual landmark. This thesis proposes a method for automatic identification of vertebral endplates by extracting the boundary using image processing. Firstly, using non-linear filters noise present in the radiographs are removed without blurring the expected edges. Segmentation of vertebral boundary is done using active contour models. The vertical component of the boundary is removed by morphological operation. The slope of horizontal component calculated from Hough transform helps to decide the vertebral endplates. The slope intercept representation of these endplates are used for quantification of Cobb angle. Scoliosis is a three dimensional deformity, only the spinal curvature is not sufficient to estimate the extent of severity. Objective estimation of vertebral rotation involves identification of apical vertebra and pedicle displacement. This thesis automates the identification of apical vertebra and position of pedicle using image processing. Scoliosis curve pattern classification are used in surgical planning. State-of-the-art classification procedures are based on manual identification of the curves at different levels as well as its deviation. This thesis extracts spinal column from the entire radiograph using customized filter and thereby avoiding misclassification.