A CUDA Accelerated Time Alignment for uMI Panorama GS
X. Lyu · 2024
Positron Emission Tomography (PET) utilizes timing information to pair two 511keV photons into a coincidence event, assuming that the timestamps assigned to each detector are synchronized. In practical systems, time delays occur due to clock skew and optical pathway length. As timing resolution improves, time calibration becomes crucial as timing offsets can quickly become the primary error source. UIH has developed the uMI Panorama GS PET/CT scanner with a 78cm ring diameter and 148.2cm axial length. Previous work introduced a time alignment method using through CPU parallel computation. CUDA, Compute Unified Device Architecture, is a parallel computing platform and programming model developed by NVIDIA. It allows developers to use NVIDIA graphics processing units (GPUs) for general-purpose processing. CUDA enables significant performance gains for computing tasks by harnessing the parallel processing power of GPUs alongside traditional CPUs. In this study, we extend the time alignment method using L1-norm minimization and CUDA acceleration, suitable for low statistics and the long axial FOV of the PET system. The wider coincidence window required by the long axial FOV leads to the necessity for a wider coincidence window during initial time calibration. We can perform fast time calibration using this method due to high sensitivity and low statistics handling capabilities of the total-body PET system. The new CUDA-based time alignment approach is suitable for this scenario. We have tested the accuracy and convergence of this method in the uMI Panorama GS PET/CT, observing improvements in time resolution.