Real-Time Closed-Loop Tracking Of An Unknown Number Of Neural Sources Using Probability Hypothesis Density Particle Filtering

Probability hypothesis density (PHD) filtering, implemented using particle filters, is a Bayesian technique used to non-linearly track multiple objects. In this paper, we propose a new approach based on PHD particle filters (PHD-PF) to automatically track the number of magnetoencephalography (MEG) neural dipole sources and their unknown states. In particular, by separating the MEG measurements using independent component analysis, PHD-PF is applied in a closed-loop to first estimate the number of sources and then recover their amplitude, location and orientation. We also reduce the processing time and computational complexity by employing window-based processing and multi-channel decomposition. We simulate the overall system using synthetic data and show that the proposed algorithm has tracking performance similar to existing approaches with significantly fewer number of particles. We also map the algorithm on to a Xilinx Virtex-5 field-programmable gate array (FPGA) platform. The processing period for one iteration using 3,200 particles is only about 314 mu s, which makes this implementation suitable for real-time tracking.