Effect of radiographic landmark identification errors on the accuracy of three-dimensional reconstruction of the human spine

Abstract

In three-dimensional reconstruction of the human spine obtained from stereo-radiographic set-ups (two radiographs or more), it is extremely difficult to identify exactly the same landmarks on all radiographs. The effect of these identification errors was investigated with simulations made on points of known three-dimensional co-ordinates and compared with three-dimensional reconstructions of real spines obtained with the direct linear transformation algorithm. Results showed that radiographic identification errors of up to 2 mm were common, causing reconstruction errors of up to 5 mm. These reconstruction errors may be noticed in the form of geometrical inaccuracies in the graphical representation of three-dimensional reconstructions of the spine. Successive displacements were then imposed on image point co-ordinates to minimise the identification error and increase the reconstruction accuracy. The improvement on the three-dimensional reconstruction results was negligible. Three-dimensional reconstructions obtained from three radiographs were also investigated. They showed slightly more accurate reconstructions than those obtained from two radiographs. However, the increase of X-ray exposure on the patient may not be worthwhile.