Optimal measurement times for particle filtering and its application in mobile tumor tracking

Particle filter is a powerful algorithm to track the state of discrete dynamic systems from noisy measurements. Under certain circumstances, the number of measurements has to be restricted. In this case, one is interested in finding the optimal measurement time set for particle filtering, either apriori, i.e. before any measurement acquisition, or online, i.e. as the measurements are acquired. These problems are referred to respectively as the apriori problem and the online problem and arise for instance, in the domain of mobile tumor tracking based on X-ray images, where X-ray acquisitions have to be made in a parsimonious way to limit patients’ exposure to harmful radiations. This work proposes to address these two problems by designing an algorithm that finds near-optimal measurement time sets. The developed algorithm is based on the nesting of a genetic algorithm, a Monte Carlo algorithm, and a particle filter. An application in mobile tumor tracking is presented and the performance of the method is measured on a simplified lung tumor model. In comparison with measurements performed at regular time intervals, the measurement time set solving the apriori problem reduces the expected mean squared tracking error by 37.5%. The expected tracking performance is improved by 39.8% using the measurement time set solving the online problem with a significantly smaller computational budget.