A probabilistic algorithm integrating source localization and noise suppression for MEG and EEG data

Abstract

We have developed a novel probabilistic model that estimates neural source activity measured by MEG and EEG data while suppressing the effect of interference and noise sources. The model estimates contributions to sensor data from evoked sources, interference sources and sensor noise using Bayesian methods and by exploiting knowledge about their timing and spatial covariance properties. Full posterior distributions are computed rather than just the MAP estimates. In simulation, the algorithm can accurately localize and estimate the time courses of several simultaneously active dipoles, with rotating or fixed orientation, at noise levels typical for averaged MEG data. The algorithm even performs reasonably at noise levels typical of an average of just a few trials. The algorithm is superior to beamforming techniques, which we show to be an approximation to our graphical model, in estimation of temporally correlated sources. Success of this algorithm using MEG data for localizing bilateral auditory cortex, low-SNR somatosensory activations, and for localizing an epileptic spike source are also demonstrated.

Introduction

Mapping the spatiotemporal neural activity of the entire brain is an important problem in basic neuroscience research. It is also clinically important for patients with brain tumors and epilepsy, both in localizing regions important for cognitive function and for identifying epileptogenic brain regions. Such brain mapping procedures are useful to guide neurosurgical planning, navigation, and resection.

Many noninvasive techniques have emerged for functional brain mapping, such as functional magnetic resonance imaging (fMRI) and magnetoencephalography (MEG). Although fMRI is the most popular method for functional brain imaging with high spatial resolution, it suffers from poor temporal resolution since it measures blood oxygenation level-dependent (BOLD) signals with fluctuations in the order of seconds. These BOLD signals are also indirect measures of neural activity which might not accurately reflect neural activity, especially in regions of altered vasculature. However, dynamic neuronal activity has fluctuations in the submillisecond time scale that can only be directly measured with magnetoencephalography (MEG) and/or electroencephalography (EEG). MEG data are the measurements of tiny magnetic fields surrounding the head while EEG data are the measurements of voltage potentials using an electrode array placed on the scalp.

The past decade has shown rapid development of whole-head MEG/EEG sensor arrays and of algorithms for reconstruction of brain source activity from MEG and EEG data, termed source localization. All existing methods for brain source localization are hampered by the many sources of noise present in MEG/EEG data. The magnitude of the stimulus-evoked neural sources is on the order of noise on a single trial, and so typically 50–200 averaged trials are needed in order to clearly distinguish the sources above noise. This limits the type of cognitive questions that can be asked and is prohibitive for examining processes such as learning that can occur over just one or several trials. Needing to average trials is time consuming and therefore difficult for a subject or patient to hold still or pay attention through the duration of the experiment. Gaussian thermal noise or Gaussian electrical noise is present at the MEG or EEG sensors themselves. Background room interference such as from powerlines and electronic equipment can be problematic. Biological noise such as heartbeat, eyeblink or other muscle artifact can also be present. Ongoing brain activity itself, including the drowsy-state alpha (∼ 10 Hz) rhythm can drown out evoked brain sources. Finally, many localization algorithms have difficulty in separating neural sources of interest that have temporally overlapping activity.

Noise in MEG and EEG data is typically reduced by a variety of preprocessing algorithms before being used by source localization algorithms. Simple forms of preprocessing include filtering out frequency bands not containing a brain signal of interest. Additionally and more recently, ICA algorithms have been used to remove artefactual components, such as eyeblinks (Jung et al., 2000). More sophisticated techniques have also recently been developed using graphical models for preprocessing prior to source localization (Nagarajan et al., 2005, Nagarajan et al., 2006). Therefore, current algorithms for source localization from MEG and EEG data typically use a two-stage procedure — the first for noise/interference removal and the second for source localization.

This paper presents a probabilistic modeling framework for MEG/EEG source localization that estimates Source Activity using Knowledge of Event Timing and Independence from Noise and Interference (SAKETINI). The framework uses a probabilistic hidden variable model that describes the observed sensor data in terms of activity from unobserved brain and interference sources. The unobserved source activities and model parameters are inferred from the data by a variational Bayesian Expectation Maximization algorithm. The algorithm then creates a spatiotemporal image of brain activity by scanning the brain, inferring the model parameters and variables from sensor data, and using them to compute the likelihood of a dipole at each grid location in the brain.

We first describe the generative model for the data. We complete specification of the model for post-stimulus sources by including prior distributions and compute the unknown quantities learned from the data. We then describe the model for learning interference and noise sources from pre-stimulus data. We finish the Methods by showing that an established source localization method, the minimum variance adaptive beamformer (MVAB) (Sekihara et al., 2001), is an approximation of our framework. In the Results section, we show performance of SAKETINI relative to MVAB and sLORETA (Pascual-Marqui, 2002) both in localization ability and in time course estimation for MEG data. We show the effect of number of sensors and time points for all three methods. We further show the proposed method's performance applied to a real auditory-evoked MEG dataset, a low-SNR somatosensory MEG dataset, and an epileptic spike MEG dataset. We conclude with a discussion on the model order and inputs to the algorithm, extensions of the model, and its relationship to other methods in the literature. A preliminary report of this work was shown in Zumer et al. (2007).