Fusion and Rendering of Volume Data

This page contains sample images that were created with an experimental system for rendering a volume data set obtained through the fusion of two orthogonal volume data sets. We explain the problem and provide some sample results.

If the images appear too dark on your monitor see the note about gamma correction.

Volume Fusion

The scanning devices for obtaining volume data typically produce a data set as a collection of images, each called a slice. A common problem with this scanning method is that a scanned volume has a higher resolution within each slice than between slices. For a CT scanner, this difference can be as high as eight to one. In the following figure,

the lines represent slices. The left data set consists of horizontal slices, and the middle data set consists of vertical slices. The fusion process merges two data sets taken at different orientations to produce a single data set, as is depicted on the right.

In this work, we explore the rendering aspect of fusion. The main difficulty here is that at a point not on either horizontal or vertical slices there is no measure data available. One way to resolve this problem is to bilinearly interpolate from nearby slices, but this approach leads to poor results when the distances between slices are larger than two. We are developing an effective approach to address the above difficulty.

MR Brain



Click on the images above to see full-size versions.

These images were rendered from a 256 x 256 x 167 MR scan of a human brain. To simulate the lower resolution between slices, we drop slices at controlled rates.

The first image is the original data set with no slices dropped.

In the second image, we take the slices in x-direction and in y-direction. After dropping the slices between every two x-slices and every two y-slices, the remaining x-slices and y-slices are merged and rendered with the traditional approach. There are some minor artifacts, but the rendering is still acceptable.

In the third image, we follow the procedure described for the second image, except now we drop the slices between every four x-slices and every four y-slices. Now the artifacts are quite serious.

In the fourth image, we drop slices as we did for the third image, but we perform the rendering with the new system. The artifacts are mostly eliminated.

MR Knee



Click on the images above to see full-size versions.

These images were rendered from a 256 x 256 x 110 MR scan of a human knee. To simulate the lower resolution between slices, we drop slices at controlled rates.

The first image is the original data set with no slices dropped.

In the second image, we take the slices in x-direction and in y-direction. After dropping the slices between every four x-slices and every four y-slices, the remaining x-slices and y-slices are merged and rendered with the traditional approach. The artifacts are serious.

In the third image, we drop slices as we did for the second image, but we perfom the rendering with the new system. The artifacts are mostly eliminated.

In the fourth image, we follow the procedure described for the second image, except now we drop the slices between every eight x-slices and every eight y-slices. Artifacts are now more apparent than those of the second image.

In the fifth image, we drop slices as we did for the fourth image, but we perfom the rendering with the new system. The artifacts are greatly reduced.

CT Head



Click on the images above to see full-size versions.

This image was rendered from a 256 x 256 x 226 CT scan of a human head. To simulate the lower resolution between slices, we drop slices at controlled rates.

The first image is the original data set with no slices dropped.

In the second image, we take the slices in x-direction and in y-direction. After dropping the slices between every two x-slices and every two y-slices, the remaining x-slices and y-slices are merged and rendered with the traditional approach. There are some minor artifacts, but the rendering is acceptable.

In the third image, we follow the procedure described for the second image, except now we drop the slices between every four x-slices and every four y-slices. Now the artifacts are serious.

In the fourth image, we drop slices as we did for the third image, but we perform the rendering with the new system. The artifacts are mostly eliminated.

People: Bede Liu, Department of Electrical Engineering, Princeton University
Acknowledgements: Many people have generously provided their help to this work. It will be a great pleasure to acknowledge these individuals when this work is complete. At that time, we will aslo augment this page with more technical details.
Last update: Dec 29, 1996
guo@dgp.toronto.edu

back to the main page