An image reconstruction method for dynamic PET

Abstract

In PET, SPECT, and other tomographic imaging modalities, dynamic image sequences are typically obtained by reconstructing the individual time frames independently. This frame-by-frame reconstruction approach can be highly suboptimal because it fails to take into account the temporal correlations in the signal. We propose an alternative approach, based on a Karhunen-Lohe transformation of the dynamic-image reconstruction problem, that can offer both significant improvement of the reconstructed images and a substantial reduction in the computation required to reconstruct an image sequence.

Multichannel Image Recovery

“…, K, is to be reconstructed. Usually the reconstruction is performed image by image, but recent research [20,21,17,15,12] has shown that it is preferable to reconstruct all of the images collectively as a multichannel image f from all of the data in the multichannel observation vector g. The following example from PET brain imaging illustrates this principle.…”