Development of Three-Dimensional Model of Spine, Using Bi-planar Radiographic Images and Adaptation of Prior CT Data

Three-dimensional reconstruction method of the thoracolumbar spine is investigated by the use of bi-planar radiographic images (lateral & posterior-anterior). In this study, a prior knowledge of the vertebrae, captured by statistical data, is used. This method has two steps: 1) determine the approximate location of each vertebra by selecting the centres of the superior & inferior endplates, and subsequently applying a rigid deformation on the statistical template of the vertebra; 2) a global deformation using a statistical modal analysis of the pathological deformations is applied to the model. An energy based function is introduced to find vertebrae deformations. Optimizing of this model is obtained with applying exploration/selection algorithm, which leads to the 3-D reconstruction of each vertebra and spinal model. This energy function depends on similarity between extracted external contours of deformed template and edge potential field of two radiographic views together with a prior energy, which is dependent on variation modes yielded from statistical knowledge. The initial results of the reconstructed models are compared with 3-D models yielded by conventional CT-scans, demonstrating good geometry fitting.