Complex Singular Value Decomposition Based Noise Reduction of Dynamic PET Images

Abstract

Individual images in dynamic molecular imaging studies are noisy because of short duration of frames. To reduce noise in these studies we used a method that employed the Hilbert transform and Singular Value Decomposition (SVD) processing. Use of this method, which we call the Complex Singular Value Decomposition (CSVD), significantly reduces noise while preserving signal intensity of dynamic images. Further, we used simulations to examine the effect of CSVD processing on estimates of receptor kinetic parameters. We found a significant reduction in variance when CSVD processing was applied to images that had Gaussian noise added to the signal. The signals were preserved even after adding noise, thus the simulations revealed that noise reduction was not at the cost of relevant signal. It therefore appears that CSVD processing can be used in dynamic molecular imaging and other similar studies to reduce noise and improve signal quality. Keywords: Singular value decomposition (SVD), complex singular value decomposition (CSVD), dynamic PET, noise reduction, molecular imaging, dopamine, raclopride, Positron Emission Tomography, SRTM