Efficient Bayesian Tracking of Multiple Sources of Neural Activity: Algorithms and Real-Time FPGA Implementation

• Published in: IEEE transactions on signal processing 2013-02, Vol.61 (3), p.633-647
• Main Authors: Lifeng Miao, Zhang, J. J., Chakrabarti, C., Papandreou-Suppappola, A.
• Format: Article
• Language: English
• Subjects: Algorithms; Applied sciences; Artificial intelligence; Bayesian analysis; Biological and medical sciences; Brain modeling; Computation; Computer science; control theory; systems; Computerized, statistical medical data processing and models in biomedicine; Connectionism. Neural networks; Detection, estimation, filtering, equalization, prediction; Dipole sources; Dipoles; Electroencephalography; Estimates; Exact sciences and technology; Field programmable gate arrays; FPGA implementation; Information, signal and communications theory; Inverse problems; Medical management aid. Diagnosis aid; Medical sciences; multiple particle filters; neural activity; parallel architecture; particle filter; Pattern recognition; probability hypothesis density filter; Real time; Real-time systems; Sensors; Signal and communications theory; Signal processing; Signal, noise; Studies; Telecommunications and information theory; Tracking; Vectors
• ISSN: 1053-587X; 1941-0476
• DOI: 10.1109/TSP.2012.2226172

We propose new Bayesian algorithms to automatically track current dipole sources of neural activity in real time. We integrate multiple particle filters to track the dynamic parameters of a known number of dipole sources, resulting in reducing the computational intensity incurred due to the large number of sensors required to observe magnetoencephalography (MEG) or electroencephalography (EEG) measurements. When we also need to estimate the time-varying number of dipole sources, we develop an algorithm based on applying probability hypothesis density filtering (PHDF) for multiple object tracking. The PHDF is implemented using particle filters (PF-PHDF), and it is applied in a closed-loop with MEG/EEG measurements to first estimate the number of sources and then their corresponding amplitude, location and orientation. The PF-PHDF tracking algorithm uses an online, window-based multiple channel decomposition processing approach that reduces the overall processing time and computational complexity. We demonstrate the improved performances of the proposed algorithms by simulating neural activity tracking systems witWe propose new Bayesian algorithms to automatically track current dipole sources of neural activity in real time. We integrate multiple particle filters to track the dynamic parameters of a known number of dipole sources, resulting in reducing the computational intensity incurred due to the large number of sensors required to observe magnetoencephalography (MEG) or electroencephalography (EEG) measurements. When we also need to estimate the time-varying number of dipole sources, we develop an algorithm based on applying probability hypothesis density filtering (PHDF) for multiple object tracking. The PHDF is implemented using particle filters (PF-PHDF), and it is applied in a closed-loop with MEG/EEG measurements to first estimate the number of sources and then their corresponding amplitude, location and orientation. The PF-PHDF tracking algorithm uses an online, window-based multiple channel decomposition processing approach that reduces the overall processing time and computational complexity. We demonstrate the improved performances of the proposed algorithms by simulating neural activity tracking systems with both synthetic and real data. We map the proposed algorithms onto Xilinx Virtex-5 field-programmable gate array (FPGA) platforms and demonstrate real-time tracking performance. For example, our results showed that the PF-PHDF algorithm can