Fast accurate stereoradiographic 3D-reconstruction of the spine using a combined geometric and statistic model Vincent Pomero a, * , David Mitton a ,S ebastien Laporte a , Jacques A. de Guise b , Wafa Skalli a a Laboratoire de Biom ecanique, ENSAM-CNRS UMR 8005, 151 Boulevard de l’H^ opital, 75013, Paris, France b Laboratoire de recherche en Imagerie et Orthop edie, CRCHUM -  ETS, Montr eal, Canada Received 10 July 2003; accepted 28 November 2003 Abstract Objective. To describe and evaluate a fast accurate stereoradiographic 3D-reconstruction method of the spine. Background. Stereoradiographic methods based on anatomical landmarks identification are the only ones providing information on 3D-deformities of the spine in a standing position, but require 2–4 h for the whole spine, making the method inadequate for clinical routine. Methods. The proposed semi-automated method is based on (1) vertebral body volume reconstruction, (2) definition of a local referential associated to this volume, (3) reliable a priori knowledge of the vertebral shape using eight morphologic descriptors of the vertebral body to estimate, from a multiple linear regression, 21 3D-point coordinates per vertebra, (4) kriging of a 2000 points model with regard to the 21 points. The method was evaluated for vertebral orientation and shape accuracy. Results. 3D models of the whole spine are obtained within 15 min. Manual vs. semi-automated reconstruction comparison yield similar accuracy regarding the CT-scan references. For vertebrae orientation, results were slightly different from the manual reconstruction method (however an absolute reference is lacking). Conclusion. The stereoradiographic 3D-reconstruction method allows for a significant reduction of the whole reconstruction time, with regard to previously described methods. Moreover, the accuracy was evaluated and was found to be comparable to the accuracy of previous methods. The results of this study show that stereoradiography could now be employed in routine clinical environment. Relevance 3D spine reconstruction from biplanar radiographs in standing position can be obtained using a fast and accurate method. Ó 2003 Elsevier Ltd. All rights reserved. Keywords: Stereoradiography; 3D-modelling; Semi-automated; Spine; Scoliosis; CT-scan; Morphological database 1. Introduction The analyses of the 3D deformities of the spine, such as scoliosis, require a 3D-reconstruction of the spinal geometry (Dubousset, 1992; Perdriolle and Vidal, 1987). In clinical practice, accurate 3D-reconstructions can be obtained using CT-scan when considering 1 mm cuts and using semi-automated segmentation techniques (Landry et al., 1997; Mitulescu et al., 2001b). Never- theless, the reconstruction of scoliotic spines by com- puted tomography (CT) imagery would require a large number of millimetric cuts which would result in an unacceptable irradiation of the patients (Levy et al., 1996). Furthermore, CT scan imaging is performed on patients in prone/supine position, which alters the real spinal curvature and therefore would bias its morpho- logical analysis. For the last 15 years, stereoradiographic protocols consisting in bi-planar radiography of different inci- dences, i.e., frontal and lateral standard radiographs, have been largely used in research environments as an alternative for 3D-reconstruction of the scoliotic spine in standing position. In order to have a geometrically * Corresponding author. E-mail addresses: vincent.pomero@paris.ensam.fr, vincent.pomero@ laposte.net (V. Pomero). 0268-0033/$ - see front matter Ó 2003 Elsevier Ltd. All rights reserved. doi:10.1016/j.clinbiomech.2003.11.014 www.elsevier.com/locate/clinbiomech Clinical Biomechanics 19 (2004) 240–247