Incorporating HYPR de-noising within iterative PET reconstruction (HYPR-OSEM)

Abstract

HighlY constrained back-PRojection (HYPR) is a post-processing de-noising technique originally developed for time-resolved magnetic resonance imaging. It has been recently applied to dynamic imaging for positron emission tomography and shown promising results. In this work, we have developed an iterative reconstruction algorithm (HYPR-OSEM) which improves the signal-to-noise ratio (SNR) in static imaging (i.e. single frame reconstruction) by incorporating HYPR de-noising directly within the ordered subsets expectation maximization (OSEM) algorithm. The proposed HYPR operator in this work operates on the target image(s) from each subset of OSEM and uses the sum of the preceding subset images as the composite which is updated every iteration. Three strategies were used to apply the HYPR operator in OSEM: (i) within the image space modeling component of the system matrix in forward-projection only, (ii) within the image space modeling component in both forward-projection and back-projection, and (iii) on the image estimate after the OSEM update for each subset thus generating three forms: (i) HYPR-F-OSEM, (ii) HYPR-FB-OSEM, and (iii) HYPR-AU-OSEM. Resolution and contrast phantom simulations with various sizes of hot and cold regions as well as experimental phantom and patient data were used to evaluate the performance of the three forms of HYPR-OSEM, and the results were compared to OSEM with and without a post reconstruction filter. It was observed that the convergence in contrast recovery coefficients (CRC) obtained from all forms of HYPR-OSEM was slower than that obtained from OSEM. Nevertheless, HYPR-OSEM improved SNR without degrading accuracy in terms of resolution and contrast. It achieved better accuracy in CRC at equivalent noise level and better precision than OSEM and better accuracy than filtered OSEM in general. In addition, HYPR-AU-OSEM has been determined to be the more effective form of HYPR-OSEM in terms of accuracy and precision based on the studies conducted in this work.

Cross-validation study between the HRRT and the PET component of the SIGNA PET/MRI system with focus on neuroimaging

“…Determined voxel noise (%) for the HRRT (Fig. 5) is comparable to already published results [28] within the limits of small differences in the phantom used and methodological approaches.…”

Spatially Compact MR-Guided Kernel EM for PET Image Reconstruction

“…For dynamic PET reconstruction, a hybridized kernel matrix [27] has also been developed, based on a similarity measure between both PET and MR feature vectors, to reduce the risk of smoothing structures unique to the PET data. More recently the kernel method has been implemented based on the HYPR denoising method [33] for dynamic PET reconstruction [34] with the kernel calculated using the total time frame reconstructed image as a composite image, and for static PET OSEM reconstruction [35] with the composite image derived from the sum of each subset’s reconstructed image at the previous iteration.…”

A Monte Carlo approach for improving transient dopamine release detection sensitivity

“…This maximized sensitivity comes with the trade-off of high CST and low parametric accuracy. Conversely, the standard method can be optimized for CST (SM-CST) by using a denoised reconstruction applying the HYPR operator after each update of OSEM, HYPR-AU-OSEM, 28 with a 2.5 mm kernel size followed by 5 mm HYPR post-processing, producing images with high parametric accuracy, but low detection sensitivity.…”