Research On BP Network-based Transfer Functions For Volume Rendering And The Applications

Nowadays direct volume rendering has been proved that it was an effective method to discover the detailed information of three-dimensional objects. The former researcher, however, committed themselves to generate the algorithms that enhancing the quality of rendering and accelerating the process of rendering, until these days the focus was drawn into the research of constructing transfer functions.The traditional one-dimensional transfer function for volume rendering only considers a voxel's scalar value; that is, there is a direct mapping of transparency to scalar value. Designing this type of transfer function is often not intuitive for the users and less effective. Recently, there has been research into the use of 2D and 3D transfer functions which take more information into account such as first- and second-order gradient. The additional derivative information allows more refined classification, and works well when a user wants to visualize the region of interest. However, other properties such as textures and position are not taken into account.The implement of the CAD in medicine practice spend a lot of research effort. CAD can help doctors to do the second aided-diagnosis, evaluate the inspection results in 3D visualization and detect the suspicious region to reduce the doctor's mistakes. Recently, a few commercial lung cancer CAD systems are present in abroad. Although the value of theory and application in these CAD systems is promising, there are many issues to address.Finding a method for specifying these transfer function in an intuitive and efficient manner remains an important challenge in making volume visualization a more attractive tool for understanding data. Higher-dimensional transfer function can lead to better results since it uses more properties for each voxel. In medical routine, the aim of visualization is highly dependent on the application. Thus, in order to achieve a successful volume classification in a reasonable period of time, part of the knowledge of an expert user must be introduced in the transfer function design process. So it's the best to combine the high-dimensional transfer function with a user interface.We present a new approach to the transfer function problem, using a BP network to abstract high-dimensional transfer function from the user. The user's interaction consists of specifying regions of interest by sampling on a number of slices from the volume data. And then we apply the transfer function to the lung cancer CAD to get a better result of classification of chest volume data, alleviating the jobs in the clinical inspection. It is instructive to the detecting of the nodules in the future work.