Reprinted with permission from the Advanced Imaging Pro article, "Researching the Brain", featured in the June 2010 issue.
Paul Villard, a French chemist and physicist, discovered gamma rays in 1900. Like most great discoveries, it happened while studying something else—in Villard’s case, radiation emitted by radium. He was leading a beam of radiation out of a shielded container of radium salts onto a photographic plate, which was shielded by a thin layer of lead. The lead stopped alpha rays, but there were two others. One was beta, one unknown. That was the gamma ray.
These waves are generated by radioactive atoms and in nuclear explosions. Gamma rays have a number of uses, but one in particular is the ability to kill living cells, which physicians take advantage of to attack certain cancers. They have the smallest wavelengths and the most energy of any wave in the electromagnetic spectrum.
Gamma rays also are being used in other medical research. An interesting project at the University of Arizona (Tucson) uses FastSPECT III, a dedicated SPECT (single photon emission computed tomography) imager designed specifically for imaging and studying neurological pathologies—including Alzheimer’s and Parkinson’s disease—in rodent brains. It incorporates 20 Point Grey Research (Richmond, B.C. Canada) Dragonfly Express cameras that stream images to Firewire 1394b framegrabbers at up to 800 Mbps or 200 frames per second at full resolution (640 x 480).
The Firewire bus is daisy chained across computers to synchronize cameras for timing purposes in dynamic imaging studies. Each computer has an Intel Core i7 quad-core processor and two Nvidia GeForce GTX 295 cards. There are two GPUs per GTX 295 card providing a total of 960 stream processors per computer. Frame parsing with multiple detectors at 200 fps is readily achieved by parallel computing using Nvidia CUDA and OpenMP APIs. Operation of the detectors at this rate corresponds to processing 4,000 fps for the entire system or 1.23 Gpix/second. Assuming uniform illumination, more than 2:5 x 104 counts per second per detector or greater than 5 x 105 for the entire system can be processed.
The researchers use what they call BazookaSPECT detector technology (so-called because of its original appearance to a Bazooka due to a long imaging chain). “It’s a whole different type of gamma-ray detector than is used in clinics,” explains Brian Miller, a PhD candidate at the University of Arizona Center for Gamma-Ray Imaging. “This differs in that it uses CCDs and has resolution orders of magnitude higher than in clinics. These are 100 microns; in the clinic, it’s 3 or 4 millimeters.”
BazookaSPECT comprises a novel combination of an image intensifier, a columnar or structured scintillation material, an optical-coupling system, and any off-the-shelf CCD/CMOS sensor. With amplification of scintillation light prior to the imaging chain, the researchers achieve great flexibility in detector design.
Twenty independent BazookaSPECT gamma-ray detectors acquire projections of a spherical field of view with pinholes selected for desired resolution and sensitivity. Each BazookaSPECT detector comprises a columnar CsI (Tl) scintillator, image-intensifier, optical lens and fast-frame-rate CCD camera operating at 200 frames per second. Data stream back to processing computers via Firewire interfaces, and heavy use of graphics processing units (GPUs) ensures that each frame of data can be processed in real time to extract the images of individual gamma-ray events.
In a white paper, the process is described as frame-parsing. “One algorithm for frame processing uses the following steps:
1. A frame from the CCD is acquired.
2. A median filter is applied to remove hot/noisy pixels.
3. The filtered image is thresholded above the noise, and individual clusters are identified via a fast connected components labeling algorithm.
4. Pixels corresponding to the identified clusters are extracted.
5. Using the associated cluster pixels, the 2D/3D interaction location is estimated optimally using maximum likelihood estimation.”
Miller says this new generation of detectors has stirred excitement in the gamma-ray imaging community, particularly in the context of small-animal SPECT. These detectors offer sub-100 μm resolutions, an order-of-magnitude higher than conventional scintillation-based gamma cameras. His responsibility includes prototype detector design and development, acquisition and processing software development, and imaging aperture design and fabrication.
“It’s the first of its kind,” he explains. “What’s novel is the image intensifier. It amplifies low light levels. At the end, it gives you [better] resolution.
“We take a radioactive atom, attach it to a protein or drug and inject it. It gets trapped in the body and is a non-invasive way of seeing. In the case of cancer, it goes to the tumor. The tumor processes the cell and the cell emits gamma rays. They’re like x-rays that come from the body. We can see if the tumor absorbs it. We have to be able to form images—radioactive forms—so we use pinholes to form the image. We block all the gamma rays except through the pinhole. Gamma rays are basically high-energy lights.”
SPECT imaging needs an array of pinholes as small as 200 microns, which are made of platinum. The holder needs to be made of a very dense metal, like lead, to block all the pinholes except those needed for a particular experiment. The Arizona lab uses tungsten and an epoxy to reach the density of lead. The custom molds are created with a 3D printer. The process allows multiple parts to be fabricated from a single mold.
In addition to small-animal SPECT imaging with the BazookaSPECT detector, Miller and his colleagues have investigated methods and applications that take advantage of the detector’s superb intrinsic resolution. This had led to development of a gamma-ray microscope based on micro-coded apertures. In pinhole imaging, ultimately the limiting factor in system resolution is the pinhole diameter. A smaller pinhole will result in higher resolution but at the expense of collection efficiency. It has been shown that near-field coded aperture imaging, with applications in small-animal SPECT, can be used to provide high-spatial resolution imaging with high collection efficiency. The team implemented near-field coded aperture techniques, scaled to utilize the resolution of BazookaSPECT, and demonstrated planar reconstructions having resolutions of 30μm; an unprecedented resolution for gamma-ray imaging. This experiment was achieved with the use of a platinum disk having 480 25 μm diameter pinholes.
This technology is another new step toward developing treatment strategies for diseases of the brain. “For example,” Miller says, “in Alzheimer’s patients, there are plaques [in the brain]. Are they the cause or the result? Drugs are coming out to remove them and see if they have any effect or not. A few years ago, this was not possible. Our acquisition system has five computers, each with graphics cards with multiple cores. They process in parallel and high resolution and in real-time frame rates. It was driven by the gaming industry. And that will drive the price down for scientists.”