Abstract
In this report, we propose a method for denoising the tissue time-activity curve (tTAC) obtained from a series of brain PET images. The tTAC is used in a various types of receptor analysis such as a linear method for estimating the total volume of distribution (VT). However, the presence of a large amount of noise in PET scanning reduces the accuracy of the estimation. In the proposed method, the measured tTAC is denoised using the parametric eigenspace of the tTAC. It is known that the tTAC is well represented by the compartment model, which is derived from the behavior of the ligand in the brain. In order to eliminate noise, we compute the eigenspace of a set of tTACs that have been artificially generated based on the compartment model. Given an observed tTAC, we project it to the eigenspace and eliminate the noise by means of MAP estimation. We evaluated the denoising performance of the proposed method using both simulated tTACs and actual clinical data. The VT was estimated using the denoised tTACs. The evaluation results showed that the bias and variance of the estimates were improved by employing our proposed denoising method. This indicates that the proposed method is useful for denoising tTACs.
