Multiple resolution PET data. Image properties using the list-mode ML-EM reconstruction algorithm

A finely pixelated silicon (Si) probe operated in coincidence with a clinical PET scanner should be able to increase the scanner's spatial resolution. For this purpose, conventional PET events, where two annihilation photons are detected in the ring, need to be combined with the higher resolution probe-ring events in such a way as to not degrade other image properties. The MADEIRA project utilizes such a probe-and-ring geometry, with the probe consisting of a stack of 10 Si detectors, of 1 x 1 x 1 mm3 pixels, in a 80 x 52 array. The system was simulated using the Monte Carlo package GATE, with a phantom that provides both hot and cold lesions for analysis. Due to the unconventional geometry of the system, the maximum likelihood-expectation maximization (ML-EM) algorithm was used to reconstruct the images. The abundance of possible measurements in the probe-ring data made it necessary to use the list mode version of the algorithm and the Siddon method was used to model the system response. Due to the varying spatial resolution of the measurements, multiple Siddon lines were used for each detector pair. Data was sorted into ringring and probe-ring events. Images were reconstructed from each data set separately and from the combined data set. Mean values in various regions of interest (ROIs) showed similar convergence in images reconstructed from the ring-ring data only and in those reconstructed from the combined data. In the ROIs with activity, mean values in the combined data are up to 7% higher than in the ring-ring images and 25% in the cold ROI, at convergence. The signal to noise ratio (SNR) and the contrast to noise ratio (CNR) reduce only slightly faster with the addition of the probe data to the conventional data. This indicates that the high-resolution data can be added to the PET ring data with little cost to contrast or other statistical image measures. (orig.)