The Splatter module developed at ARSC performs real-time volume rendering of large 3D data sets, such as environmental or MRI data [8]. Using the parallel processing power of the T3D, it is possible to render large volumetric datasets interactively at a rate of several images per second, compared to minutes or longer per image on other machines. Such a renderer opens the possibility of creating powerful new learning and diagnosis tools using high performance computing and networking. For example, interactive MRI images could form the basis for a virtual reality dissection laboratory. The availability of such a tool on the national network would also allow medical specialists to perform analyses and consultations on high resolution MRI images which would otherwise be impossible. This level of visualization complexity can only be achieved with the fast generation of data provided by the T3D.
Figure 9: T3D Splatter module in AVS network.
The Y-MP and T3D processes for the Splatter module are both written in C. The T3D process employs shmem message passing and communicates with the Y-MP via process memory files. The Splatter module is shown in an AVS network in Figure 9. The output from the module is an image which can be displayed using the image viewer or display tracker modules. The input port on the Splatter module accepts a camera transformation matrix from the display tracker module, which allows the 3D view to be controlled interactively.
Figure 10 is an example of a volume rendered CT image from the Visible Human dataset[9]. Bone material is white, muscle is red and softer tissues appear in yellow. The Visible Human dataset from which this image was created consisted of 1800 slices with a resolution of 512 by 512 voxels per slice. At four bytes per voxel, the complete head-to-toe dataset is about 1.8GB in size.
Figure 10: Visible Human image rendered by Splatter module on T3D.
The Splatter module has a multilevel control panel which allows interactive control of all 3D volume viewing parameters, such as lighting, shading, color and transparency. When Splatter is used with the display tracker module, the view can be rotated, scaled and translated interactively using the AVS mouse conventions for viewing 3D objects. Using 64 PEs on the T3D, the time to render the entire 1.8GB Visible Human volume dataset from a new point of view is about 5 seconds.