Abstract
In this paper, we explore the multiple source localisation problem in the cerebral cortex using magnetoencephalography (MEG) data. We model neural currents as point-wise dipolar sources which dynamically evolve over time, then model dipole dynamics using a probabilistic state space model in which dipole locations are strictly constrained to lie within the cortex. Based on the proposed models, we develop a Bayesian particle filtering algorithm for localisation of both known and unknown numbers of dipoles. The algorithm consists of a region of interest (ROI) estimation step for initial dipole number estimation, a Gibbs multiple particle filter (GMPF) step for individual dipole state estimation, and a selection criterion step for selecting the final estimates. The estimated results from the ROI estimation are used to adaptively adjust particle filter's sample size to reduce the overall computational cost. The numerical results show that the proposed methods can achieve improved performance metrics in terms of dipole number estimation and dipole localisation.
Introduction
Brain source localization is a critical challenge in non-invasive neuroimaging. The goal is to determine the location and number of electrical currents (dipoles) in the cerebral cortex based on magnetic fields measured outside the head via magnetoencephalography (MEG). This task is an "ill-posed inverse problem" because multiple internal configurations can produce identical external magnetic fields.
Current methodologies often struggle with two primary aspects: dynamic tracking of sources that change intensity and position over milliseconds, and identifying an unknown number of active sources. Many existing approaches rely on oversimplified spherical head models or rigid discrete grids, which fail to capture the continuous geometry of the brain's surface, leading to reduced accuracy or high computational demands.
Methodology
The proposed framework, the Gibbs Multiple Particle Filter (GMPF), addresses these challenges through three key innovations:
Continuous Realistic Head Model
Rather than using a simple sphere, the method employs a 1-layer realistic head model (the Nolte model). By applying nearest-neighbor (NN) interpolation, the algorithm creates a smooth manifold that constrains dipole dynamics to the complex geometry of the cerebral cortex. This ensures that the estimated sources remain physically plausible within the brain's structure.
Gibbs Sampling Integration
To account for the statistical coupling between sources that are physically close, the GMPF incorporates a Gibbs sampling step. While standard filters often treat sources as independent to reduce complexity, Gibbs sampling iteratively updates each dipole's state conditional on the others, allowing the filter to disentangle overlapping magnetic signatures.
Adaptive Filtering and ROI Estimation
The algorithm begins with a "Region of Interest" (ROI) estimation using noise-normalized Minimum Norm Estimation (MNE) to provide an initial count of dipoles. As the particle filter converges, it adaptively reduces the sample size. This adaptive scheme significantly lowers computational overhead without compromising localization precision.
Results and Discussion
The performance of the GMPF was evaluated using synthetic MEG data for a 204-magnetometer system, allowing for comparison against a known ground truth.
Localization Accuracy
In scenarios with three active dipoles, the continuous model achieved a Root Mean Square Error (RMSE) of approximately 5.6 mm. This is a significant improvement over standard SIR filters (~12.7 mm) and discrete grid models (~9.3 mm). The continuous interpolation allows dipoles to move naturally between grid points, capturing finer spatial details.
Computational Efficiency
The adaptive particle reduction scheme proved highly effective. The GMPF completed a 50-step iteration in 28 seconds, nearly twice as fast as the standard Multiple Particle Filter (MPF), which required 56 seconds. This efficiency makes the method more viable for processing large datasets.
Dynamic Source Tracking
The model successfully tracked the number of active dipoles even during "birth-death" processes where sources appeared and disappeared over time. The GMPF followed the ground truth more closely than traditional ROI-based methods, demonstrating robustness in dynamic environments.
Conclusion
The Bayesian particle filtering method presented here offers a robust solution for brain source localization. By integrating a realistic continuous head model with adaptive Gibbs sampling, the GMPF improves both accuracy and speed. Future work will focus on validating the algorithm with real clinical data and exploring parallel computing techniques to scale the method for a larger number of simultaneous dipoles.
Summary updated