Fast 3D Reconstruction of the Spine Using User-Defined Splines and a Statistical Articulated Model

This paper proposes a method for rapidly reconstructing 3D models of the spine from two planar radiographs. For performing 3D reconstructions, users only have to identify on each radiograph a spline that represents the spine midline. Then, a statistical articulated model of the spine is deformed until it best fits these splines. The articulated model used on this method not only models vertebrae geometry, but their relative location and orientation as well. The method was tested on 14 radiographic exams of patients for which reconstructions of the spine using a manual identification method where available. Using simulated splines, errors of 2.2±1.3mm were obtained on endplates location, and 4.1±2.1mm on pedicles. Reconstructions by non-expert users show average reconstruction times of 1.5min, and mean errors of 3.4mm for endplates and 4.8mm for pedicles. These results suggest that the proposed method can be used to reconstruct the human spine in 3D when user interactions have to be minimised.