Virtual Endoscopy

Over the last 5 years, virtual endoscopy has emerged as a new imaging and visualization technique to explore the human anatomy. It is probably best described by Jolesz et al. in [85]:
``Endoscopy is used to view the inner surfaces of hollow organs in a continuous fashion using optical, video-assisted technology. By changing the position of the endoscope, the operator can view the inside of an organ while controlling the viewing position and angle of the probe. [...]
Endoscopy yields detailed anatomy of the inner surfaces of a displayed wall segment [...] The method provides a direct, relatively high resolution views and, in most cases, obtains access through natural orifices or small incisions (i.e., laparoscopy). Nonetheless, endoscopic procedures can be uncomfortable and sedation or anesthesia may be required. Furthermore, endoscopes display only the inner surface of hollow organs and yield no information about the anatomy within or beyond the wall. This limitation prevents evaluation of the transmural extent of tumors and limits the ability to localize the lesion relative to surrounding anatomic structures.
CT and MRI provide cross-sectional images in which the inner surfaces of hollow organs are displayed at a much lower resolution than by endoscopy, but the techniques are noninvasive. Conventional slice-by-slice presentation of these data precludes contiguous viewing of the inner wall, forcing the radiologist to create a mental picture of anatomic continuity. This slice-based visual inspection is quite difficult and may be faulty, especially with highly convoluted tubular organs like small bowel or tortuous blood vessels. Traditional computer integration of cross-sectional data into 3D renderings have provided only outer surfaces of organs, which in the case of hollow or tubular structures are diagnostically less important. Despite these disadvantages, volumetric CT and MRI data can provide information not accessible by the endoscope. These important features include: information on tissue extending through and beyond organ walls, and the anatomic context of the entire volume, which permits correct localization of the lesion in relationship to adjacent anatomic structures.

  

Figure: Virtual bronchoscopy. Left: 3d model and camera position. Right: endoscopic view. (image from Jolesz et al. [85])

Virtual endoscopic or fly-through methods [85,84,65,171,111,172] which combine the features of endoscopic viewing and cross-sectional volumetric imaging may provide an advance in diagnosis, however, so far there are no comparative clinical data to demonstrate the advantages of virtual endoscopy. Nevertheless, virtual endoscopic presentation of image data enables the operator not only to explore the inner wall surfaces but also to navigate inside the virtual organs extracted from CT or MR images. [...]''
Because of these promising features, virtual endoscopy has been applied to many different organs, including bronchoscopy [171,111], colonoscopy [172], pancreatoscopy [113], laryngoscopy [58], sinus endoscopy [67] or otoscopy [57,90].
The extension of the technique to such different areas requires an appropriate choice of 3D imaging technique - Computed Tomography (CT), MR Imaging or a combination of those - , of the contrast agents, ... The image volume is then subject to a number of image processing tools in order to create a 3D model of the organ one wants to visualize. This model can then be either volume [134] or surface [97] rendered, the later allowing faster (interactive) visualization. For instance, in figure , the surface of the model was generated using the marching cubes algorithm [96].
After the 3D model of the organ has been created, the last critical step is the selection of the camera's position. Jolesz et al. [85] propose three techniques to guide the virtual camera through the surface models:

• Manual camera movement, where the mouse controls the camera position and focal point interactively. For instance, figure shows the graphical user interface used by Frankenthaler et al. [57].
  
    

  Figure: Graphical user interface for the interactive camera control. (Image from Frankenthaler et al. [57])

  
  
  
• Key-framing, that interpolates between user-selected key points, using cubic splines to calculate intermediate parameter values at any desired resolution. This technique was used to generate motion through open interior and exterior environments.
• Automatic path planning, a technique adapted from Lengyel's robot path planning algorithm [93]. The camera is considered as a point robot and the walls of the organs as obstacles. The path planner labels all voxels with a distance to the terminus of the organ to be explored. Given a starting point, the path planning algorithm finds the shortest path to the goal using a steepest descent algorithm.

Automatic path planning is particularly useful for tubular organs, that present a challenge for both manual camera movement and key-framing. Indeed, manual camera movement is particularly difficult within confined spaces.
Unfortunately, the automatic path planning technique suffers from several imperfections. Lengyel's algorithm uses the city-block metric in the geodesic distance computation. This restricts the path to directions along the axis, which produces an unpleasantly jagged camera movement. In the order to compensate this effect, the path is smoothed. This can in turn introduce a new adverse effect, when the camera path crosses the object boundaries and includes positions outside of the organ.
This chapter proposes methods to improve the quality of automatic path planning. In section , we show how the geodesic DT algorithm of chapter 8 can be adapted to compute the shortest path between two points. In section , we propose a new smoothing technique based on the snake model, that searches a trade-off between the smoothness of the path and its centering in the middle of the bowel.