Graphics processing units (GPUs) are increasingly used for general purpose calculations. Their pipelined architecture
can be exploited to accelerate various parallelizable algorithms. Medical imaging applications are
inherently well suited to benefit from the development of GPU-based computational platforms. We evaluate in
this work the potential of GPUs to improve the execution speed of two common medical imaging tasks, namely
Fourier transforms and tomographic reconstructions. A two-dimensional fast Fourier transform (FFT) algorithm
was GPU-implemented and compared, in terms of execution speed, to two popular CPU-based FFT routines.
Similarly, the Feldkamp, David and Kress (FDK) algorithm for cone-beam tomographic reconstruction was implemented
on the GPU and its performance compared to a CPU version. Different reconstruction strategies
were employed to assess the performance of various GPU memory layouts. For the specific hardware used, GPU
implementations of the FFT were up to 20 times faster than their CPU counterparts, but slower than highly
optimized CPU versions of the algorithm. Tomographic reconstructions were faster on the GPU by a factor up to
30, allowing 256<sup>3</sup> voxel reconstructions of 256 projections in about 20 seconds. Overall, GPUs are an attractive
alternative to other imaging-dedicated computing hardware like application-specific integrated circuits (ASICs)
and field programmable gate arrays (FPGAs) in terms of cost, simplicity and versatility. With the development
of simpler language extensions and programming interfaces, GPUs are likely to become essential tools in medical
Single photon emission computed tomography (SPECT) is an important technology for molecular imaging studies of small animals with an increasing demand for high performance imaging systems. We have designed a small animal imaging system based on position sensitive avalanche photodiodes (PSAPDs) detectors with the goal of submillimeter spatial resolution and high detection efficiency, which will allow us to minimize the radiation dose to the animal and to shorten the time needed for the imaging study. Our design will use fourteen 80×80 mm<sup>2</sup> PSAPD detectors, which can achieve an intrinsic spatial resolution of 0.5 mm. These detectors are arranged in two rings around the object and are equipped with pinhole collimators to produce magnified projection data. A mouse bed is positioned in the center of the detector rings and can be rocked about the central axis to increase angular sampling of the object. The performance of this imaging system and of a dual headed SPECT system has been simulated using a ray tracing program taking into account appropriate point spread functions. Projection data of a hot rod phantom with 84 angular samples have been simulated. Appropriate Poisson noise has been added to the data to simulate an acquisition time of 15 min and an activity of 18.5 MBq distributed in the phantom. Both sets of data were reconstructed with an ML-EM reconstruction algorithm. We also derived spatial resolution and detection efficiency from analytical equations and compared the performance of our system to a variety of other small animal SPECT imaging systems. Simulations show that our proposed system produces a spatial resolution of 0.9 mm which is in good agreement with the resolution derived from analytical equations. In contrast, simulations of the dual headed SPECT system produce a spatial resolution of 1.1 mm. In comparison to other small animal SPECT systems, our design will offer a detection efficiency which is at least 2-fold higher at better or comparable spatial resolution. These results suggest that detectors based on PSAPD technology can be used to improve the design of small animal SPECT imaging systems considerably. Our small animal system design is very compact and can achieve high resolution and detection efficiency.
We have developed a method that uses a large amount of <i>a priori</i> information to generate super resolution radiographs. We measured and modeled analytically the point spread function of a low-dose gas microstrip x-ray detector at several beam energies. We measured the relationship between the local image intensity and the noise variance in the radiographs. The soft-tissue signal in the images was modeled using a minimum-curvature filtering technique. These results were then combined into an image deconvolution procedure using wavelet filtering to reduce restoration noise while keeping the enhanced small-scale features. The method was applied to a resolution grid image to measure its effects on the detector’s modulation transfer function. The restored images of a radiological human-torso phantom revealed small-scale details on the bones that were not seen before, and this, with improved SNR and image contrast. Dual-energy imaging was integrated to the restoration process in order to generate separate high-resolution images of the bones, the soft tissues, and the mean atomic number. This information could be used to detect bone micro-fractures in athletes and to assess bone demineralization in seniors due to osteoporosis. Super resolution radiographs are easier to segment due to their enhanced contrasts and uniform backgrounds; the boundaries of the features of interest can be delimited with a sub-pixel accuracy. This is highly relevant to the morphometric analysis of complex bone structures like individual vertebrae. The restoration method can be automated for a clinical environment use.