High-fidelity visualization of large medical datasets on commodity hardware
A novel rendering method aimed at providing medical-quality rendering of large medical datasets using an ordinary desktop PC. GPU-based volume rendering cannot provide high-fidelity images when datasets do not fit into the graphics memory, whereas CPU-based volume rendering cannot provide an adequate frame rate when large datasets are used.
When using GPU-based volume ray casting for medical data visualization on consumer-level computers, a problem arises: the amount of VRAM memory available. Modern imaging devices can acquire a very large amount of data, often in an order of magnitude of gigabytes. Common graphics cards have 256MB, at most 512MB, of memory available onboard, so they are not able to store large medical datasets, the dimensions of which can exceed the gigabyte. A common solution to this problem is to subsample the data, reducing its extent so as to fit it into the VRAM memory. However, this approach drastically decreases the quality of the image, and is not suitable for medical data visualization, since anatomical parts can be amenable to visualization artifacts which in turn can lead to wrong diagnoses.
Therefore, GPU-based volume ray casting algorithms allow us to visualize large volumetric datasets at interactive frame rates due to the extremely parallel nature of direct volume rendering, but at the cost of a lower resolution of the final image. On the contrary, CPUs can store the entire dataset inside the RAM memory, so CPU-based volume ray casting algorithms are able to produce high resolution images but at the cost of a lower frame rate.
The hybrid approach takes advantage of the heterogeneity of the resources available on off-the-shelf computers: the large availability of system memory and the parallel-oriented architecture of modern GPUs. The CPU is used to render in high resolution a region-of-interest of the volume, whereas the GPU renders the context using a downsampled version of the dataset.
The rendering method has been implemented as an extension of VTK, a widely used open source library, to promote its use and evolution.
Hybrid CPU-GPU VR
Last update: 19/03/2013
|Windows 7 / Vista / XP 32 bit||5.9 MB|
|Windows 7 / Vista / XP 64 bit||6.4 MB|
Hybrid CPU-GPU VR is an open source software released under the terms of the GNU General Public License.
The source code of both the extension of the VTK library and the application will be released soon.
- L. Gallo and A. P. Placitelli, "High-fidelity visualization of large medical datasets on commodity hardware", submitted.