Global Sources
EE Times-Asia
Stay in touch with EE Times Asia
EE Times-Asia > Embedded

Speeding up medical imaging process using FPGA

Posted: 01 Sep 2011 ?? ?Print Version ?Bookmark and Share

Keywords:Medical imaging? PET scanner? FPGA? synthesis tool?

Medical imaging tasks can call for high-performance signal processing to convert sensor data into imagery to help with medical diagnostics. FPGAs are an ideal platform for these systems, since they can perform heavily pipelined operations customized to the exact needs of a given computation. In a previous work, we have benchmarked a CT scanner back-projection algorithm. In this article we focus on an FPGA platform and a high level synthesis tool called Impulse C to speed up a statistical line of reaction (LOR) estimation for a high-resolution Positron Emission Tomography (PET) scanner. The estimation algorithm provides a significant improvement over conventional methods, but the execution time is too long to be practical for clinic applications. Impulse C allows us to rapidly map a C program into a platform with a host processor and an FPGA coprocessor. In this article, we describe some successful optimization methods for the algorithm using Impulse C. The results show that the FPGA implementation can obtain an 82x speedup over the optimized software.

PET is a nuclear medical imaging technique which produces a three-dimensional scan of the body. This technology is used in hospitals to diagnose patients and in laboratories to image small animals. In the beginning of the process, radiotracers are injected into bodies and absorbed by specific tissues. A radiotracer is a molecule labeled with a positron emitter, such as 11C, 13N, 18F or 15O. In the nucleus of a radiotracer, a proton decays and a positron is emitted. After traveling a short distance, the emitted positron annihilates with an electron and produces two 511 keV photons which travel in essentially opposite directions.

 PET scanner

Figure 1: Shown is the inaccuracy of conventional PET scanner. The estimation error is cause by ignoring the depth and order of interaction.

In order to determine the distribution of radiotracers, a patient is placed between abundant gamma detectors. When a pair of photons is detected within a short timing window, the system registers the event as a coincidence. The line joining two relevant detectors is called a line of response (LOR). A gamma detector is composed of crystals that interact with photons in order to improve the signal to noise ratio. However, it is difficult to estimate the exact depth of interactions. Moreover, a photon may interact with multiple crystals, which makes it harder to estimate the true LOR. Conventional PET scanners employ versions of Anger logic to position events [1][2]. Anger logic positions interactions by calculating the energy-weighted centroid of the measured signals. The depth of interaction is assigned a constant value which is based on the attenuation coefficient of the detector crystals. It causes the well known parallax error [2] (figure 1). Therefore, it is important for a high resolution PET scanner to estimate the depth and order of interactions.

Methods and tools
In order to estimate the first interaction positions of photons, a statistical LOR estimation method is proposed in [3]. It uses a Bayesian estimator for determining a LOR by estimating three-dimensional interaction positions of a pair of annihilation photons. This estimation algorithm provides a significant improvement over Anger logic under a variety of parameters, but it also involves huge computations. The optimized algorithm takes 1.76 hours to estimate LORs from one million coincidences running in a computer with an AMD Opteron Processor 168. It requires processing millions of LORs in order to construct a high-resolution image visualizing the distribution of the radiotracers. For example in [4], it needs 250 million LORs for reconstructing an image. The execution time for the algorithm to construct such a high-resolution image will be 18.3 compute-days. This is unacceptable for clinic applications.

Our goal is to reduce the execution time of the LOR estimation algorithm by using Field-Programmable Gate Arrays (FPGA). FPGA devices can provide high computation power and flexibility. Traditional FPGA implementations use hardware description languages (HDLs) such as VHDL and Verilog. HDL programming methodologies aimed at chip designs are unsuitable for programming large-scale scientific applications [5]. In this project, we use a C-to-FPGA tool-flow called Impulse C. Impulse C allows us to rapidly map a C program into a platform such that the FPGA can co-process with a host processor. In [6] we showed that the design time using Impulse C is less than the HDLs. Moreover, a program written in Impulse C is portable. Without modifying any code it can be re-mapped into a newer FPGA device or platform.

 PET system

Figure 2: Here's a sketch of the PET system and the dMiCE crystal pair.

We sped up the LOR estimation of a PET scanner using this high level synthesis approach. On the XtremeData XD2000F [13] platform, the system estimates LORs from one million coincidences in 79 seconds at 100MHz. A speedup of 82x is achieved compared to the optimized software. Another contribution is that we describe some methods for improving the application.

The algorithm
The algorithm was developed for a high-resolution PET detector capable of providing depth-of-interaction information. Our collaborators at U.W. designed the PET system as an insert to a mammography unit; figure 2 shows the sketch of the PET system. There are four detector panels in the system, and each detector-panel is composed of depth-of-interaction micro-crystal element (dMiCE) crystal pairs. The size of the individual crystal element is 2220 mm3 and each is coupled to its own 22 mm2 micro-pixel avalanche photodiodes (MAPDs) [7]. A triangular-shaped light reflector is placed between two crystal elements in a dMiCE pair. The differential distribution of the scintillation light shared can be used to determine the depth of interaction.

1???2???3???4?Next Page?Last Page

Article Comments - Speeding up medical imaging process ...
*? You can enter [0] more charecters.
*Verify code:


Visit Asia Webinars to learn about the latest in technology and get practical design tips.

Back to Top