Dynamic PET imaging using the Kernel Method
Dynamic PET imaging monitors the spatiotemporal distribution of a radiotracer in the human body in vivo and has the ability to quantitatively characterize the radiotracer kinetics of the underlying molecular process. This requires the tracer activity within each image voxel to be accurately measured during multiple short time frames - which, however, suffers from high noise due to low counts that can be acquired in short scan durations. The kernel-based dynamic (KBD) image reconstruction has been recently proposed by our group to conquer this high-noise problem in dynamic PET. KBD improves image reconstruction of low-count data by incorporating prior information derived from high-count composite data of many time frames using machine learning. The method has been shown to provide a stark improvement to the quality of dynamic PET activity images. This presentation will focus on the evaluation of a new kernel type which has been designed to improve region of interest (ROI) quantification in dynamic PET images.