A simple and robust method for fast crystal identification

In Positron Emission Tomography (PET), crystal look-up tables (CLUT) are used to map the positions of incident annihilation photons to individual crystals in a typical 2D scintillator detector. To obtain the CLUT, a flood image of the detector is first acquired by irradiating the detector with a flood source. The flood image is then segmented into regions, each corresponding to a specific crystal in the detector. In this paper, a simple and robust method is proposed for fast crystal identification. First, morphology methods and a Gaussian kernel are applied to enhance the contrast of the image. Local maxima is then searched and identified as peaks. A fast method is developed to sort the peaks and the borders of the crystal are calculated using the coordinates of adjacent peaks. CLUT is then generated based on the borders of the crystals. A MATLAB graphical user interface (GUI) tool incorporating the proposed method is also developed to facilitate optional interactive operations. The method was tested using (1) a flood image of an Inveon detector; (2) four flood images of PET insert detectors; (3) a flood image of a clinical PET scanner detector; (4) a flood image of a prototype PETbox detector and (5) four flood images of MicroPET R4 detectors. For all the flood images tested, 3 edge crystals in one of the MicroPET R4 detectors and 8 edge crystals in the prototype PETbox detector were failed to be identified. Besides that, all the crystals were automatically segmented and the resulting CLUT were reasonably satisfactory. The computational time for segmentation in the MATLAB environment was less than 2 sec for each flood image on a standard Pc. Results show that the proposed method exhibits reasonably good performance as well as robustness dealing with different flood images and we believe it could be used in the daily basis where thousands of crystals need to be accurately identified.