At the University of Erlangen in Germany, some thorough research is going on with regard to the rapid developments in medical imaging. The academic experts, Markus Blank, Ralf Petzold and Willi Kalender have given an account of the progressive improvements in visual data rendering which finally have resulted in virtual endoscopy (VE). VE presents itself as a powerful tool that might partly substitute the invasive process of conventional endoscopy in the near future. The so-called fly through procedure allows the physician to conduct a more effective examination providing better comfort for both patient and operator.
Medical imaging modalities such as spiral computed tomography (CT) and magnetic resonance (MR) permit the supply of high-resolution volumetric data sets. These have to be adapted for optimal display on a computer screen, a process called visualisation which helps the physician to build his diagnosis and to plan surgery. The simplest method for visualisation constitutes multiplanar reformation (MPR) and consists of displaying planar cuts through the volumetric data set. Although advanced MPR techniques are able to vary the thickness of the cuts and even allow curved cuts, the medical observer has to be very experienced to mentally reconstruct a three-dimensional picture out of two-dimensional cuts.
The shaded surface displays (SSD) approach is a more complex technique which extracts the object of interest out of the volume by means of thresholding in order to create a geometric meta object to which lighting is applied during the projection onto the screen. This implies a data reduction speeding up the rotation and translation of the object. It is possible to introduce perspective during projection, as if the physician would stand right inside the object and look at the inner surface. SSDs are particularly used to visualise dense tissue with a smooth surface. If mathematical rays are cast through the data volume, the maximum intensity projection (MIP) technique is applied which selects the maximal intensity encountered along each ray path to construct the image. Contrarily to SSD which only projects the surface data, MIP needs all the volume data to compute the projection of specifically the object's dimensions.
Just like MIP, volume rendering (VR) casts parallel projection rays through the volume but uses an opacity function connecting every intensity with an additional opacity value to create the image. VR thus allows the physician to look from the outside at any location within the volume while obscuring structures are made transparent or semitransparent. Like with SSD, perspective can be added through the use of divergent projecting rays. This perspective volume rendering (pVR) technique makes close objects appear larger than objects of identical size at a greater distance. Objects, which otherwise would be hidden, can be visualised by placing internal viewpoints beyond obscuring structures.
Physicians apply the pVR method to explore tubular structures in the human body in order to view all the information simultaneously instead of in a single two-dimensionally way. They even can fly through this scene and turn pVR into virtual endoscopy if they produce continuous views of the same structure to display them in rapid sequence. Key frames are in advance being defined along the flight path to save time while the remaining gaps are filled up afterwards during the animation creation process. Since VE is not invasive nor limited in movement and is able to reach any place in the volume by driving through obstacles, physicians prefer this method to conventional mechanical endoscopy.
At present, the computational visual rendering and the actual VE performance are still very time consuming and demand a great deal of experience since interaction is yet impossible. Paradoxically, virtual endoscopy allows the physician to save time and money in medical examination. Although VE constitutes no tool for surgical intervention, it serves perfectly well as a means for surgery planning or as a surgery guide. Yet, further development may lead to other applications as you can read in the article Visible Human Datasets form ideal basis for "endoscopic virtuality", also available in this issue. More technical details about the current visualisation techniques are to be found at the University of Erlangen web site.