Multitarget data association with higher-order motion models for tracking in proton CT instrumentation

Abstract

PRaVDA is an advanced imaging and dosimetry system for proton therapy with solid-state sensors throughout. The residual energy resolving detector (range telescope) can be composed of multiple layers of CMOS imagers. Such imagers, though operating at relatively low frame rates, can record many protons events per read cycle. The challenge is to identify the non-linear tracks of individual protons through the telescope. A multi-target data association with higher-order motion model (MDAMM) was designed for the proton events in each layer. Such algorithms focus more on motion constraints rather than appearance. We use a cost function based on Gaussian for MDAMM which is suitable for protons as they are laterally scattered. We factor in all trajectories of a protons path (through all layers) into a cost function in order to improve the quality of a candidate trajectory. Also to achieve a more efficiency and better target tracking for protons, each trajectory with its previously created tracking result is considered in the cost function.