Apply to real-time medical imaging of eye stereoscopic imaging system

"We assess for parallel computation of a variety of development environments," responsible for CUDA implementation of researchers Herlambang so said, "finally chose the CUDA, because it allows us to use the familiar C-language syntax to open challenges

In Imaging, a very interesting area is eye stereoscopic imaging technology that requires no special glasses will be able to display three-dimensional images.

This interesting technology not only has the potential of entertainment, can also be used as a variety of professional applications and practical technology. The University of Tokyo Graduate School of information science and technology of mechanical information Takeyoshi Dohi Professor and his colleagues studied the NVDIA of CUDA ™ parallel computing platform that medical imaging is this platform very promising application areas.

Since 2000, the University's research team has developed a system, by CT or MRI scan, real-time access to live sections are considered in texture, this system is not only able to reproduce by volume rendering as a three-dimensional image, also serve as stereo video displays for the IV system.

The system is a real-time, 3D, in Vivo Imaging brings revolutionary change.

However, it is a huge amount of computation, volume rendering itself leads to extremely high processing workload, then also need to be further processed to enable stereo imaging. For each image frame, there are many angles required to display both. This is multiplied by the number of frames in the video, you will see the shocking number of huge calculation, and must be in a very short time to finish so highly accurate calculations.

In the 2001 study, using a Pentium III 800 MHz PC to handle the number of 512 x 512 resolution of pictures, real-time volume rendering and 3D reconstruction takes 10 seconds or more time to generate a frame.

In order to expedite processing, the research team is trying to use is equipped with 60 of UltraSPARC III CPU block 900 MHz machine, which was the highest performance of the computer. But you can get the best results it is five frames per second. From a practical point of view, this speed is not fast enough.

Solution

Volume rendering and subsequently converted to IV format data parallel vector calculation.

To this end, the best computing paradigm is a GPU. Accordingly, Liao and study using CUDA Herlambang started realization of GPU, it is common from NVIDIA GPU C language development environment.

First, researchers are using the latest generation of GPU GeForce ® 8800 GTX has developed a prototype system.

When using CUDA GPU Institute 2001, running on a data set, the performance boost to 13 to 14 per second. UltraSPARC systems cost up to tens of millions of yen, is 100 times the GPU, and the GPU has delivered almost equivalent to the performance of his three-fold, researchers feel very surprised. Moreover, according to the Group's research, NVIDIA's GPU is the latest multi-core CPU at least 70 times faster. In addition, the test shows that for larger scale of the texture data, GPU performance.

Currently, this research group is now using NVDIA recent desktop-side supercomputers Tesla ™ CUDA D870, against the use of Tesla optimization present IV system.

This initiative is expected to make a significant performance gain.

Effect

In addition, we also do not have to modify the system has been developed on the premise of using faster new generation GPU.

If a large-scale environment enables debug CUDA programs possible, CUDA will surely become a more powerful parallel computing development environment, we want it in the field of medical imaging are more widely used. ”

If the ability to view real-time stereo mode from CT and MRI for image, a doctor will be able to check the status of patient organizations and make diagnosis without living check and surgical treatment.

In addition, some doctors can simultaneously view the images, communicate with each other. This allows some doctors can simultaneously arthroscopic surgery and other minimally invasive surgical techniques, and the type of each surgeon can real-time observation of surgical procedures.

The massive parallel computing array into clinical device very difficult, but Tesla GPU and a powerful computing capacity allows for compact parallel computing modules possible.

Integral Videography (IV) principle sketch map

　　

Photo: computer calculation of internal processing IV element image IV pixel figure calculated Voxel data virtual three-dimensional data lens array virtual lens arrays

Flat flat panel display

Space image formation: spatial stereo image composition

Micro lens array: micro-convex lens array observer: Observer

Solution

Distance study IV image sample.

Before the display

The location of two meters form the very realistic, three-dimensional image of yellow bamboo, looks like is being held in the Palm of your hand.

Even when the observer moves, the image remains "is in the hands of" visible state. To create a high resolution, 3D images, using volume rendering methods such as, but requires extremely high computing power.

　　

Effect

Use CUDA's IV system

enter>