| In medical imaging such as Single Photon Emission Computed Tomography (SPECT), the point spread function (PSF) of the system is spatially variant. The current method of compensation is to model the PSF in the iterative reconstruction process of the image itself. However, this can be quite computationally expensive and may require a large amount of time to process on readily available computers at non-research hospitals. Possible approaches to speeding up this process have focused on using clusters of parallel computers or high speed graphic boards. Rather than simply relying on increases in computational power, this paper instead proposes a postprocessing method to compensate for the spatially variant PSF. First, the raw reconstructed image (without PSF compensation and thus requiring much less computational burden to compute) is further-blurred with a spatially variant kernel in order to produce an image with a spatially invariant PSF. This further-blurred image is then filtered using an efficient filter to compensate for the now spatially invariant PSF. This breaks the problem down into a two-step process, the first being a forward problem and the second being a conventional filtering process that can be accomplished by known methods, most often in the Fourier domain (such as a Wiener filter). This new approach is compared to the current state-of-the-art iterative ordered subset expectation maximization (OS-EM) reconstructive method by deblurring heart phantom data using each method. Two imaging setups are utilized to gather two different data sets; one utilizes a Low-Energy High-Sensitivity (LEHS) collimator with a wider acceptance angle, while the other utilizes a Low-Energy High-Resolution (LEHR) collimator with a narrower acceptance angle. It is shown that in the LEHS study, the OS-EM method takes approximately 46 hours whereas the new further-blurring then deblurring method takes approximately 14 minutes, while in the LEHR study, the OS-EM method takes approximately 20 minutes whereas the new further-blurring then deblurring method takes approximately 5 minutes. This is an efficient and cost-effective approach, especially for wider acceptance angle collimators, and the decreased computational time of the postprocessing method will allow high quality images to be used real-time in nonresearch hospital settings. |
