An expectation-maximization method for spatio-temporal blind source separation using an AR-MOG source model.

In this paper, we develop a maximum-likelihood (ML) spatio-temporal blind source separation (BSS) algorithm, where the temporal dependencies are explained by assuming that each source is an autoregressive (AR) process and the distribution of the associated independent identically distributed (i.i.d.) innovations process is described using a mixture of Gaussians. Unlike most ML methods, the proposed algorithm takes into account both spatial and temporal information, optimization is performed using the expectation-maximization (EM) method, the source model is adapted to maximize the likelihood, and the update equations have a simple, analytical form. The proposed method, which we refer to as autoregressive mixture of Gaussians (AR-MOG), outperforms nine other methods for artificial mixtures of real audio. We also show results for using AR-MOG to extract the fetal cardiac signal from real magnetocardiographic (MCG) data.

Performance comparison of six independent components analysis algorithms for fetal signal extraction from real fMCG data.

In this study we compare the performance of six independent components analysis (ICA) algorithms on 16 real fetal magnetocardiographic (fMCG) datasets for the application of extracting the fetal cardiac signal. We also compare the extraction results for real data with the results previously obtained for synthetic data. The six ICA algorithms are FastICA, CubICA, JADE, Infomax, MRMI-SIG and TDSEP. The results obtained using real fMCG data indicate that the FastICA method consistently outperforms the others in regard to separation quality and that the performance of an ICA method that uses temporal information suffers in the presence of noise. These two results confirm the previous results obtained using synthetic fMCG data. There were also two notable differences between the studies based on real and synthetic data. The differences are that all six ICA algorithms are independent of gestational age and sensor dimensionality for synthetic data, but depend on gestational age and sensor dimensionality for real data. It is possible to explain these differences by assuming that the number of point sources needed to completely explain the data is larger than the dimensionality used in the ICA extraction.

A novel adaptive beamformer for MEG source reconstruction effective when large background brain activities exist.

This paper proposes a novel prewhitening eigenspace beamformer suitable for magnetoencephalogram (MEG) source reconstruction when large background brain activities exist. The prerequisite for the method is that control-state measurements, which contain only the contributions from the background interference, be available, and that the covariance matrix of the background interference can be obtained from such control-state measurements. The proposed method then uses this interference covariance matrix to remove the influence of the interference in the reconstruction obtained from the target measurements. A numerical example, as well as applications to two types of MEG data, demonstrates the effectiveness of the proposed method.

Feature extraction using information-theoretic learning.

A classification system typically consists of both a feature extractor (preprocessor) and a classifier. These two components can be trained either independently or simultaneously. The former option has an implementation advantage since the extractor need only be trained once for use with any classifier, whereas the latter has an advantage since it can be used to minimize classification error directly. Certain criteria, such as Minimum Classification Error, are better suited for simultaneous training, whereas other criteria, such as Mutual Information, are amenable for training the feature extractor either independently or simultaneously. Herein, an information-theoretic criterion is introduced and is evaluated for training the extractor independently of the classifier. The proposed method uses nonparametric estimation of Renyi's entropy to train the extractor by maximizing an approximation of the mutual information between the class labels and the output of the feature extractor. The evaluations show that the proposed method, even though it uses independent training, performs at least as well as three feature extraction methods that train the extractor and classifier simultaneously.

RSVP Keyboard: An EEG Based Typing Interface.

Humans need communication. The desire to communicate remains one of the primary issues for people with locked-in syndrome (LIS). While many assistive and augmentative communication systems that use various physiological signals are available commercially, the need is not satisfactorily met. Brain interfaces, in particular, those that utilize event related potentials (ERP) in electroencephalography (EEG) to detect the intent of a person noninvasively, are emerging as a promising communication interface to meet this need where existing options are insufficient. Existing brain interfaces for typing use many repetitions of the visual stimuli in order to increase accuracy at the cost of speed. However, speed is also crucial and is an integral portion of peer-to-peer communication; a message that is not delivered timely often looses its importance. Consequently, we utilize rapid serial visual presentation (RSVP) in conjunction with language models in order to assist letter selection during the brain-typing process with the final goal of developing a system that achieves high accuracy and speed simultaneously. This paper presents initial results from the RSVP Keyboard system that is under development. These initial results on healthy and locked-in subjects show that single-trial or few-trial accurate letter selection may be possible with the RSVP Keyboard paradigm.

Context Information Significantly Improves Brain Computer Interface Performance - a Case Study on Text Entry Using a Language Model Assisted BCI.

We present recent results on the design of the RSVP Keyboard - a brain computer interface (BCI) for expressive language generation for functionally locked-in individuals using rapid serial visual presentation of letters or other symbols such as icons. The proposed BCI design tightly incorporates probabilistic contextual information obtained from a language model into the single or multi-trial event related potential (ERP) decision mechanism. This tight fusion of contextual information with instantaneous and independent brain activity is demonstrated to potentially improve accuracy in a dramatic manner. Specifically, a simple regularized discriminant single-trial ERP classifier's performance can be increased from a naive baseline of 75% to 98% in a 28-symbol alphabet operating at 5% false ERP detection rate. We also demonstrate results which show that trained healthy subjects can achieve real-time typing accuracies over 90% mostly relying on single-trial ERP evidence when supplemented with a rudimentary n-gram language model. Further discussion and preliminary results include our initial efforts involving a locked-in individual and our efforts to train him to improve his skill in performing the task.

Fusion with language models improves spelling accuracy for ERP-based brain computer interface spellers.

Event related potentials (ERP) corresponding to a stimulus in electroencephalography (EEG) can be used to detect the intent of a person for brain computer interfaces (BCI). This paradigm is widely utilized to build letter-by-letter text input systems using BCI. Nevertheless using a BCI-typewriter depending only on EEG responses will not be sufficiently accurate for single-trial operation in general, and existing systems utilize many-trial schemes to achieve accuracy at the cost of speed. Hence incorporation of a language model based prior or additional evidence is vital to improve accuracy and speed. In this paper, we study the effects of Bayesian fusion of an n-gram language model with a regularized discriminant analysis ERP detector for EEG-based BCIs. The letter classification accuracies are rigorously evaluated for varying language model orders as well as number of ERP-inducing trials. The results demonstrate that the language models contribute significantly to letter classification accuracy. Specifically, we find that a BCI-speller supported by a 4-gram language model may achieve the same performance using 3-trial ERP classification for the initial letters of the words and using single trial ERP classification for the subsequent ones. Overall, fusion of evidence from EEG and language models yields a significant opportunity to increase the word rate of a BCI based typing system.

Optimal set of EEG electrodes for rapid serial visual presentation.

In our application, the goal is to search through a large image to find all instances of a pre-specified, high-valued target. One approach taken to increase the throughput of this image search task is to: chop the large image into numerous small images, display them to a user at high rates one-at-a-time, and then search the simultaneously-recorded EEG data for neural activity that signifies that the user detected an instance of the target. The temporal efficiency of this EEG-based system is reduced by the overhead, which increases as the number of electrodes increases. Hence, we wish to find a minimal set of electrodes that ideally maintains the detection performance. In order to inform the design of future EEG-based image search systems, in this paper we find the 12 out of 32/64 most important electrodes for detection using 5 different feature selection methods. The optimal set includes all 5 occipital and the 2 most frontal electrodes.

Source localization of EEG/MEG data by correlating columns of ICA and lead field matrices.

Independent components analysis (ICA) has previously been used to denoise EEG/magnetoencephalography (MEG) signals before performing neural source localization. Source localization is then performed using a method such as beamforming or dipole fitting. Here we show how ICA can also be used as a source localization method, negating the need for beamforming and dipole fitting. This type of approach is valid whenever an estimate of the forward (mixing) model for all putative source locations is available, which includes EEG and MEG applications. The proposed method consists of estimating the forward model using the laws of physics, estimating a second forward model using ICA, and then correlating the columns of the matrices that represent the two forward models. We show that, when synthetic data are used, the proposed localization method produces a smaller localization error than several alternatives. We also show localization results for real auditory-evoked MEG data.

A probabilistic algorithm for robust interference suppression in bioelectromagnetic sensor data.

Magnetoencephalography (MEG) and electroencephalography (EEG) sensor measurements are often contaminated by several interferences such as background activity from outside the regions of interest, by biological and non-biological artifacts, and by sensor noise. Here, we introduce a probabilistic graphical model and inference algorithm based on variational-Bayes expectation-maximization for estimation of activity of interest through interference suppression. The algorithm exploits the fact that electromagnetic recording data can often be partitioned into baseline periods, when only interferences are present, and active time periods, when activity of interest is present in addition to interferences. This algorithm is found to be robust and efficient and significantly superior to many other existing approaches on real and simulated data.

A graphical model for estimating stimulus-evoked brain responses from magnetoencephalography data with large background brain activity.

This paper formulates a novel probabilistic graphical model for noisy stimulus-evoked MEG and EEG sensor data obtained in the presence of large background brain activity. The model describes the observed data in terms of unobserved evoked and background factors with additive sensor noise. We present an expectation maximization (EM) algorithm that estimates the model parameters from data. Using the model, the algorithm cleans the stimulus-evoked data by removing interference from background factors and noise artifacts and separates those data into contributions from independent factors. We demonstrate on real and simulated data that the algorithm outperforms benchmark methods for denoising and separation. We also show that the algorithm improves the performance of localization with beamforming algorithms.

Minimax mutual information approach for independent component analysis.

Minimum output mutual information is regarded as a natural criterion for independent component analysis (ICA) and is used as the performance measure in many ICA algorithms. Two common approaches in information-theoretic ICA algorithms are minimum mutual information and maximum output entropy approaches. In the former approach, we substitute some form of probability density function (pdf) estimate into the mutual information expression, and in the latter we incorporate the source pdf assumption in the algorithm through the use of nonlinearities matched to the corresponding cumulative density functions (cdf). Alternative solutions to ICA use higher-order cumulant-based optimization criteria, which are related to either one of these approaches through truncated series approximations for densities. In this article, we propose a new ICA algorithm motivated by the maximum entropy principle (for estimating signal distributions). The optimality criterion is the minimum output mutual information, where the estimated pdfs are from the exponential family and are approximate solutions to a constrained entropy maximization problem. This approach yields an upper bound for the actual mutual information of the output signals - hence, the name minimax mutual information ICA algorithm. In addition, we demonstrate that for a specific selection of the constraint functions in the maximum entropy density estimation procedure, the algorithm relates strongly to ICA methods using higher-order cumulants.

Epileptic seizure detection from ECoG using recurrence time statistics.

A recurrence time statistics T1 is defined and used as a feature extraction method for seizure detection. The preliminary data shows that during seizure T1 generates a peak and this peak clearly distinguishes the seizure state from background activity. When applied to multi-channel ECoG recordings, the spatial-temporal signature of T1 can be clearly observed to discriminate seizures. The T1 feature was used for automated seizure detection on two sets of long term monitoring ECoG data. The detection probability reached 97% with a 0.29 per hour average false alarm rate.

Evaluation of a BSS algorithm for artifacts rejection in epileptic seizure detection.

A data efficient blind sources separation (BSS) algorithm has been applied to preprocess intracranial EEG (ECoG) for artifact rejection. After artifacts correction a recurrence time statistics T1 feature was evaluated from the 'cleaned' data. Seizure detection performance was compared between BSS preprocessing and without preprocessing. Test results show that in a data set, for a detection rate of 96%, the false alarm rate dropped from 0.13 per hour without BSS preprocessing to 0.08 with preprocessing. For the other set of data, the false alarm rate dropped from 0.34 to 0.21 at a detection rate of 100%.