Simultaneous spatiotemporal tracking of multiple positron sources using spectral clustering

Tracking point-like objects in optically opaque environments has applications in studying flow in chemical systems, monitoring respiratory motion, and possibly investigating the trafficking of single cell at the whole-body level. One approach is to label the point-like object (particle, fiducial marker, or cell) with positron-emitting tracers. The information contained in the list-mode measurements can be utilized to directly track the motion of a single source, either frame-by-frame or continuously using a trajectory reconstruction algorithm. Applying this concept to multiple simultaneously moving sources is more challenging, in that detected pairs of coincident annihilation photons from different sources are not directly distinguishable. To tackle this problem, we applied spectral clustering methods to cluster the list-mode data into groups of coincidence lines corresponding to distinct sources. With this method, each individual source can then be independently tracked using existing techniques. Advantaged of this method are that the number of clusters can be extracted from the data, and that the method operates directly on the list-mode data, without needing to backproject the CLs into the spatial domain.