Improving PET-based physiological quantification through methods of wavelet denoising.

Abstract

The goal of this study was to evaluate methods of multidimensional wavelet denoising on restoring the fidelity of biological signals hidden within dynamic positron emission tomography (PET) images. A reduction of noise within pixels, between adjacent regions, and time-serial frames was achieved via redundant multiscale representations. In analyzing dynamic PET data of healthy volunteers, a multiscale method improved the estimate-to-error ratio of flows fivefold without loss of detail. This technique also maintained accuracy of flow estimates in comparison with the "gold standard," using dynamic PET with O15-water. In addition, in studies of coronary disease patients, flow patterns were preserved and infarcted regions were well differentiated from normal regions. The results show that a wavelet-based noise-suppression method produced reliable approximations of salient underlying signals and led to an accurate quantification of myocardial perfusion. The described protocol can be generalized to other temporal biomedical imaging modalities including functional magnetic resonance imaging and ultrasound.