A flexible approach to motion correction in nuclear medicine

Motion correction of the abdominal-thoracic region is one of the main research challenges in tomographic nuclear medicine imaging. We address this issue with a flexible data-driven method of motion correction. This uses marker-less stereo tracking of the anterior abdominal-chest surface and a ´virtual dissection´-based registration approach, combined within a novel paricle filtering (PF) framework. The key advantage to this data driven approach is that we do not make gross prior assumptions on the configuration of the hidden state of the system, i.e. the configuration of the internal organs during the emission acquisition process. Instead, at some given time instance, we infer the hidden or unobserved internal organ configuration by using Monte Carlo sampling (and then filtering) of various propositions, or ´particles´. Such estimates are calculated using the previous state (of the internal organs) plus some noise or perturbation of the expected transition to the current state or configuration. We then compare estimated representations of the abdominal-chest anterior surface, derived from the particles or propositions, with an observation of the actual surface, derived from a marker-less stereo imaging system. By examining the differences between the estimated particle or proposition surfaces and actual observed surface data, we can infer the current configuration of the internal organs. After an update step, the process is then repeated for subsequent time points in the emission data. This allows the system to flexibly adopt previously unknown configurations of the internal organs, and thus allow for different modes of breathing (e.g. abdominal vs thoracic-based motion) to be represented. Preliminary results are presented based on using the XCAT phantom to demonstrate the PF approach and the ´virtual-dissection´ registration process, alongside results of a parameterized anterior surface model derived from human volunteer data.