Pragmatic image reconstruction for high-resolution PET scanners

In this study, a pragmatic approach to image reconstruction was presented for data from high resolution positron emission tomography (PET) scanners. The proposed approach is modeled on fully-3D image reconstruction used in clinical PET scanners, which is based on Fourier rebinning (FORE) followed by 2D iterative image reconstruction. The use of iterative methods allows modeling physical effects, while FORE accelerates the reconstruction process by reducing the fully-3D data to a stacked set of 2D sinograms.; Prior to the image reconstruction study, an analytical simulation model was developed to investigate how system design parameters affect image figures of merit for mouse-imaging PET scanners. For a high resolution imaging system, important physical effects that impact image quality are positron range, annihilation photon acollinearity, detector point-spread function and coincident photon count levels (i.e., statistical noise). Modeling of these effects was included in an analytical simulation that generated multiple realizations of sinograms with varying levels of each effect. To evaluate image quality with respect to quantitation and detection task performance, four different figures of merit were measured: (1) root mean square error; (2) a region of interest SNR (SNR_ROI); (3) non-prewhitening matched filter SNR (SNR_NPW); and (4) recovery coefficient. The results indicate that positron range and non-stationary detector point-spread response effects, which are ignored for clinical imaging, cause significant reductions of quantitation (SNR_ROI) and detection (SNR_NPW) accuracy for small regions.; The proposed pragmatic image reconstruction utilized the results from the aforementioned quantitative study by applying the important physical effects on image quality to iterative reconstruction methods. To model the effects of positron range and non-stationary detector point-spread response in the proposed reconstruction method, we have added a factorized system matrix to the ASPIRE reconstruction library. The current implementation uses FORE+AWOSEM followed by post-reconstruction 3D Gaussian smoothing. The results were presented by point source simulation data from the four-ring MiCES target system and by measured data from the single ring MiCES evaluation system both of which are currently under development at the University of Washington. The results indicate that the proposed approach produces a dramatic improvement in resolution without undue increases in noise.