Correction methods for missing data in sinograms of the HRRT PET scanner

The High Resolution Research Tomograph (HRRT) is a 3D PET scanner designed for brain imaging and small animal imaging. The HRRT consists of 8 panel detector heads that are separated by a gap of 17 mm resulting in data gaps in the sinogram. Furthermore, data gaps can result from detector-block failure. To prevent artifacts in the reconstruction when using FORE, filling in of the data gaps is required. The purpose of this study was to evaluate the accuracy of several gap filling methods. Two gap-filling methods were investigated: a) bilinear interpolation, b) a model-based method: an intermediate volume is reconstructed (2D) based on only direct planes, after which this image is forward projected towards the data gaps. In addition, an improved model-based method is introduced: c) first fill the gaps using interpolation, then reconstructing using FORE and forward projecting to fill the gaps. Detector gaps and block failures were mimicked by zeroing LORs in simulated and experimentally acquired sinograms. The gaps were filled using the different methods, reconstructed using FORE+2DOSEM and compared with reconstruction of the original sinogram. From the variance of the reconstructions and from difference images it could be concluded that for homogeneous objects which are large as compared to the extent of data gaps all methods give similar results, although the interpolation methods requires significant less computation time. For objects with dimensions comparable to the size of a data gap the interpolation method falls short. The simple model-based method however suffers from artifacts in the intermediate direct planes reconstruction. The latter is overcome by the improved model-based method. In conclusion, the improved model-based method might outperform the interpolation method, but due to the long computation times usage of this method is only justified in case of small objects.