Compression of Dynamic PET Based on Principal Component Analysis and JPEG 2000 in Sinogram Domain

A new algorithm for the compression of dynamic positron emission tomography (PET) data is presented. It consists of a temporal compression stage based on the application of principal component analysis (PCA) directly to the PET sinograms to reduce the dimensionality of the data. This is followed by a spatial compression stage using JPEG 2000 to each PCA channel weighted by the signal in each channel. By combining these temporal and spatial compression techniques we can achieve a compression ratio as high as 129:1 while simultaneously reducing noise and improving functional estimation compared with the uncompressed data, and preserving the sinogram data for later analysis. We validate our approach with a simulated phantom FDG brain study and clinical dynamic PET datasets. The results of performance evaluation suggest the new compression technique not only is able to reduce the original sinogram datasets by more than 95%, but also improve the reconstructed image quality for the quantitative analysis.