PCA filter for szintigraphic imaging

Quality of image data in nuclear medicine diagnostics is mainly limited by radioactive dose and acquisition time, both yielding low total counts. There are studies with a maximum pixel count of 10 in a frame, thus signal-to-noise-ratio (SNR) is very poor. A novel data driven method was developed to enhance SNR but preserving underlying image features. In contrast to traditional filtering methods calculation of smoothed image data is not based to a region surrounding the pixel, but image structures are separated from noise using a simple statistical model. The signal is reconstructed from principal components images (PCI) and the amount of noise is assessed by chi² statistics. With this method even fine structures are preserved. The potential of the novel technique is demonstrated with in vivo data from perfusion and Parkinson studies. After removal of heavy noise, enhanced image are subject to sophisticated image processing, as automated inter-modality image registration.