Robust motion correction for respiratory gated PET/CT using weighted averaging

Movement degrades image quality in PET/CT. A common strategy is to gate the PET data, reconstruct the images, register each image to a reference gate, and average the registered images (Reconstruct, Registered, Average or RRA). Previously, we have shown that this technique can provide more quantitatively accurate PET images compared to the ungated case, while using all the data means that RRA images have better counting statistics than single gates. However, in some instances, e.g. large patient motion, low count statistics acquisitions and/or mismatch between PET and CT position, the registration might fail for a particular gate in some region of the image, resulting in artifacts in the final motion corrected image. We have developed techniques to automatically detect such failures for every image voxel based on an analysis of the motion vectors and the image values. We propose to use a weighted average, where the weights are gate and voxel dependent, and are determined based on estimated registration quality. We illustrate this idea using respiratory gated PET data with PET-PET registration using the level-sets non-rigid registration algorithm. We applied this technique to respiratory gated phantom and patient data from GE Discovery PET/CT scanners. The results demonstrate that the method reliably mitigates any image artifacts and is robust to large lesion displacements and low count density PET images.