Detection of motion in SPECT using multi-head data combination

An algorithm to automatically detect and eliminate motion in projection data from triple head SPECT systems is proposed. The approach relies on a specific protocol in which each camera rotates 3 times faster than in conventional acquisitions and performs a full 360-degree orbit, instead of the usual partial 120 degrees orbit. This permits to obtain three full sets of data. The hypothesis is that by adequately combining the sets, it is possible in most eases to reconstitute a motion-free set of projection data. The motion detection consists of searching consistent frames through the study by cross-correlating the projection sets as a function of the angle. Plateaus of high correlation are estimated and matched in order to identify projections corresponding to the same position of the subject throughout the study. Different motion situations were created by manually moving a cardiac phantom at different times. In all cases, it was possible to reconstitute the evolution of the motion during the study. When no motion occurred during an interval equal to at least one third of the total acquisition time, high correlation plateaus could be identified. It was then possible to build an optimal complete set of consistent frames from the longest matching plateaus regardless of the frequency and magnitude of motion. Otherwise, an averaging of the three sets was performed, which still improved the quality of the reconstructed slices compared to conventional imaging protocol