A statistical image-based approach for the 3D reconstruction of the scoliotic spine from biplanar radiographs

Abstract

In this paper, we propose a hybrid approach using a statistical 3D model of the spine generated from a database of 732 scoliotic patients with high-level anatomical primitives identified and matched on biplanar radiographic images for the three-dimensional reconstruction of the scoliotic spine. The 3D scoliotic curve reconstructed from a coronal and sagittal radiograph is used to generate an approximate statistical model based on a transformation algorithm which incorporates intuitive geometrical properties. An iterative optimization procedure integrating similarity measures such as deformable vertebral contours and epipolar constraints is then applied to globally refine the 3D anatomical landmarks on each vertebra level of the spine. A qualitative evaluation of the retro-projection of the vertebral contours obtained from the proposed method gave promising results while the quantitative comparison yield similar accuracy on the localization of low-level primitives such as the six landmarks identified by an expert on each vertebra.