Goals GoalsMRI is a powerful noninvasive method for imaging anatomical structures, for measuring blood oxygenation changes induced by neuronal activation and for assessing cellular metabolism. For several years a technique known as functional MRI (fMRI) is used to do "brain-mapping" experiments. A fMRI system consists of a clinical MRI-scanner with enhanced image transfer capabilities and a data analysis/display software package to aquire, reconstruct, transfer, and analyze signal responses within the repetition time period. For each newly acquired dataset, the display should be updated rapidly. The goal is a real-time system which uses a high-speed network to link an MRI-scanner with a supercomputer to make it possible to convert scan data almost instantaneously into an animated 3-D image showing what parts of the brain "light up" during mental activity. This is a complex metacomputing scenario, where imaging hardware, a massively parallel computer and a powerful visualization server at different locations need to cooperate and to rapidely exchange huge amounts of data over a wide area network.
Functional analysisA widespread method to detect brain activity is the correlation analysis where the computation of the correlation coefficient between the model of the expected signal change (reference vector) and the time series of the considered voxel is realized. The computation for each voxel is independent of the neighboring voxels and can be performed in parallel. The software package is based on a client/server model. The real-time server gets the data of the MRI-scanner and transfers them to a client running on the CRAY T3E at FZJ. A message protocol based on TCP is used for communication. After the correlation analysis the original anatomical image and the functional result are transferred to an SGI visualization server at IMK.VMSD at GMD.
This picture shows a 2-D slice of the original anatomical image on the left hand side. On the right hand side, a color overlay with the functional result (correlation) is presented.
VisualizationThe visualization of anatomical and functional data is performed on an Onyx 2 visualization server which has four independent graphical subsystems called pipes and is located at GMD. The raw data arriving from the T3E at Jülich is imported directly from the network into AVANGO (the Virtual Reality operating system developed by GMD IMK.VMSD) via a specialized network-import module. After the raw data slices (slices because they are MRI cuttings through the brain) are loaded into the main memory they are preprocessed by color coding and matching. Now a volume data set is generated from the slices and used to visualize the brain which was the original source of the data. As shown in the following illustration, 3-D textures with a billboard algorithm are used:
The figures show two anatomical slices of a human brain and a 3-D model created from 128 of such slices. The volume data set is rendered as a stereo image. This means that two independent pictures, one for each eye, are produced which can be viewed with shutter glasses (shown below) to get a 3-D impression of the brain.
The stereo images can be displayed on a fast framebuffer or a Workbench, currently at GMD, but eventually at FZJ. In order to perform an interactive control over the visualization, a device like the Polhemus Stylus (shown above) has to be connected to the framebuffer and both video data and control data from the Stylus have to be exchanged between the framebuffer and the visualization server. With this configuration, the image of the head can be translated, rotated, zoomed and cut. It is worth mentioning that the cutting plane does not have to be in the same direction as the MRI recording. Hence it is possible to cut the head in any direction and to zoom in selected regions of interest.
StatusThe parallelization of the correlation algorithm is currently under way. Importing, preprocessing, and rendering of the anatomical data sets is implemented. The resulting 3D image can interactively be visualized on a Workbench at GMD.
