RSM3D technology: 3D from very few radiographic projections

Objectives: To present the state-of-the-art of the rsm3d-technology developed in our work group. RSM3D enables 3D reconstructions from only very few, almost arbitrarily positioned radiographic projections.

Methods: Using different human specimen (skull, entire mandible, mandible segment, teeth), rsm3d reconstructions were obtained from eight to 11 projections. Projection radiographs were exposed with different image receptors (ccd-sensor, large area amorphous selenium detector, storage phosphor plates). Using the newly developed rsm3d software-tool, reconstructions were computed by means of algebraic methods on the Graphics Processing Unit (GPU).

Results: Without a visible loss in quality, reconstruction time on GPU when compared to CPU (Central Processing Unit) was reduced by a factor of 32 to 65. Typical few-view artifacts were partly suppressed by implementation of constraints such as the minimization of the total variation norm.

Conclusions: RSM3D is one approach to generate 3D information from only very few projections. Our in vitro results seem promising, however the technology is not yet ready for clinical tests. Currently, a prototype device for preclinical application for (partial) skull-reconstructions is developed.