ICA separation of functional components from dynamic cardiac PET data

The aim of this study was to improve detection of different heart tissues, and specially their boundaries, in H₂^{1 5}O PET (positron emission tomography) heart images. This problem was considered as a blind source separation problem. In order to solve it we applied ICA (independent component analysis) on dynamic image data and measured projection profiles (sinograms). The testing was based on two kinds of data: a simple dynamic numerical phantom and human heart data acquired during resting state. The sensitivity of ICA to noise was examined on phantom data, where ICA seemed to be less sensitive to noise on sinogram data than on image data. On cardiac rest data, the results were in line with the results on phantom data.