Introduction

To treat diseases, injuries and congenital or acquired deformities of the head and neck, maxillo-facial surgeons deal with complex surgery. For example, the correction of disfiguring facial birth defects requires the manipulation of scull bones with maximum precision. The pre-operative simulation of such procedures requires a 3D computer model of the patient's face. We describe an approach to create such a 3D patient model by ultra-fast holographic recording and automatic scan matching of synchronously captured holograms. The pulsed hologram records the patient's portrait within a single laser shot (pulse duration appr. 35 ns). This so-called master-hologram contains the complete 3D spatial information which, due to the extremely short recording time, is not affected by involuntary patient movements.

In a second step, the real image of the hologram is optically reconstructed with a cw-laser. By moving a diffusor-screen through the real image, a series of 2D images is projected and digitized with a CCD camera. This process is referred to as hologram tomography[3]. Each projection shows the surface contour of the patient where the image is in focus. The method was first introduced as the locus of focus technique[8] in the context of non-medical imaging. Beside the desired intensity from in-focus points from the object contour, each captured image also contains a blurred background of defocused parts of the real image. The main problem of locating the surface is therefore to distinguish between focused and unfocused regions in each slice. This procedure yields a relief map of the visible (as seen from the hologram) parts of the patient. In order to record a complete 360$^\circ$ model of a patient, multiple holograms are recorded synchronously, i.e., with the same laser pulse.

Subsequently, the resulting relief maps are registered (i.e. their relative orientation is calculated) by automated scan matching. The problem of automated scan matching is to find a transformation, consisting of a rotation and a translation, that minimizes a cost function that contains the Euclidian distances between points pairs[6] which both represent the same surface shape. Given that the surface shapes are acquired independently from locus-of-focus analysis of two synchronously recorded holograms, such an approach yields 360$^\circ$ models of complex surfaces.