Theta mediated dynamics of human hippocampal-neocortical learning systems in memory formation and retrieval.

Episodic memory arises as a function of dynamic interactions between the hippocampus and the neocortex, yet the mechanisms have remained elusive. Here, using human intracranial recordings during a mnemonic discrimination task, we report that 4-5 Hz (theta) power is differentially recruited during discrimination vs. overgeneralization, and its phase supports hippocampal-neocortical when memories are being formed and correctly retrieved. Interactions were largely bidirectional, with small but significant net directional biases; a hippocampus-to-neocortex bias during acquisition of new information that was subsequently correctly discriminated, and a neocortex-to-hippocampus bias during accurate discrimination of new stimuli from similar previously learned stimuli. The 4-5 Hz rhythm may facilitate the initial stages of information acquisition by neocortex during learning and the recall of stored information from cortex during retrieval. Future work should further probe these dynamics across different types of tasks and stimuli and computational models may need to be expanded accordingly to accommodate these findings.

A Scalable Open-Set ECG Identification System Based on Compressed CNNs.

Deep learning (DL) is known for its excellence in feature learning and its ability to deliver high-accuracy results. Its application to ECG biometric recognition has received increasing interest but is also accompanied by several deficiencies. In this study, we focus on applying DL, especially convolutional neural networks (CNNs), to ECG biometric identification to address these deficiencies. Using prestored user-specific feature vectors, the proposed scheme can exclude unregistered subjects to realize "open-set" identification. With the help of its scalable structure and "transfer learning," new subjects can be enrolled in an existing system without the need for storing the ECGs of those previously enrolled. Finally, schemes based on the quantum evolutionary algorithm (QEA) are presented to prune unnecessary filters in the proposed CNN model. The performance of the proposed scheme was evaluated using the ECGs of 285 subjects from the PTB dataset. The experimental results demonstrate an identification rate of more than 99% in closed-set identification. Although incorporating the proposed method for unregistered subject exclusion degraded the identification performance slightly, the ability of the approach to resist a dictionary attack was evident. Finally, using the QEA-based filter pruning method and its two-stage extension reduced the number of floating-point operations required to complete one identity recognition to 1.20% and 0.22% of the original value without significantly impacting the identification accuracy.

Drosophila Brain Functional Data Analysis: A Unified Framework.

A unified framework for the analysis of fluorescence data taken by a two-photon imaging system is presented. As in the processing of blood-oxygen-level-dependent signals of functional magnetic resonance imaging, the acquired functional images have to be co-registered with a structural brain atlas before delineating the regions activated by a given stimulus. The voxels whose calcium traces are highly correlated with the predicted responses are demarcated without the need for subjective reasoning. Experimental data acquired while presenting olfactory stimuli are used to demonstrate the efficacy of the proposed schemes. The results indicate that the functional images of a Drosophila individual can be normalized into a standard stereotactic space, and the expected brain regions can be delineated adequately. This framework provides an opportunity to enable the development of a Drosophila functional connectome database.

A novel framework for feature extraction in multi-sensor action potential sorting.

BACKGROUND: Extracellular recordings of multi-unit neural activity have become indispensable in neuroscience research. The analysis of the recordings begins with the detection of the action potentials (APs), followed by a classification step where each AP is associated with a given neural source. A feature extraction step is required prior to classification in order to reduce the dimensionality of the data and the impact of noise, allowing source clustering algorithms to work more efficiently. NEW METHOD: In this paper, we propose a novel framework for multi-sensor AP feature extraction based on the so-called Matched Subspace Detector (MSD), which is shown to be a natural generalization of standard single-sensor algorithms. RESULTS: Clustering using both simulated data and real AP recordings taken in the locust antennal lobe demonstrates that the proposed approach yields features that are discriminatory and lead to promising results. COMPARISON WITH EXISTING METHOD(S): Unlike existing methods, the proposed algorithm finds joint spatio-temporal feature vectors that match the dominant subspace observed in the two-dimensional data without needs for a forward propagation model and AP templates. CONCLUSIONS: The proposed MSD approach provides more discriminatory features for unsupervised AP sorting applications.

Matching pursuit and source deflation for sparse EEG/MEG dipole moment estimation.

In this paper, we propose novel matching pursuit (MP)-based algorithms for EEG/MEG dipole source localization and parameter estimation for multiple measurement vectors with constant sparsity. The algorithms combine the ideas of MP for sparse signal recovery and source deflation, as employed in estimation via alternating projections. The source-deflated matching pursuit (SDMP) approach mitigates the problem of residual interference inherent in sequential MP-based methods or recursively applied (RAP)-MUSIC. Furthermore, unlike prior methods based on alternating projection, SDMP allows one to efficiently estimate the dipole orientation in addition to its location. Simulations show that the proposed algorithms outperform existing techniques under various conditions, including those with highly correlated sources. Results using real EEG data from auditory experiments are also presented to illustrate the performance of these algorithms.

Projection versus prewhitening for EEG interference suppression.

Suppression of strong, spatially correlated background interference is a challenge associated with electroencephalography (EEG) source localization problems. The most common way of dealing with such interference is through the use of a prewhitening transformation based on an estimate of the covariance of the interference plus noise. This approach is based on strong assumptions regarding temporal stationarity of the data, which do not commonly hold in EEG applications. In addition, prewhitening cannot typically be implemented directly due to ill conditioning of the covariance matrix, and ad hoc regularization is often necessary. Using both simulation examples and experiments involving real EEG data with auditory evoked responses, we demonstrate that a straightforward interference projection method is significantly more robust than prewhitening for EEG source localization.

Efficient dipole parameter estimation in EEG systems with near-ML performance.

Source signals that have strong temporal correlation can pose a challenge for high-resolution EEG source localization algorithms. In this paper, we present two methods that are able to accurately locate highly correlated sources in situations where other high-resolution methods such as multiple signal classification and linearly constrained minimum variance beamforming fail. These methods are based on approximations to the optimal maximum likelihood (ML) approach, but offer significant computational advantages over ML when estimates of the equivalent EEG dipole orientation and moment are required in addition to the source location. The first method uses a two-stage approach in which localization is performed assuming an unstructured dipole moment model, and then the dipole orientation is obtained by using these estimates in a second step. The second method is based on the use of the noise subspace fitting concept, and has been shown to provide performance that is asymptotically equivalent to the direct ML method. Both techniques lead to a considerably simpler optimization than ML since the estimation of the source locations and dipole moments is decoupled. Examples using data from simulations and auditory experiments are presented to illustrate the performance of the algorithms.

Performance bounds for dynamic causal modeling of brain connectivity.

The use of complex dynamical models have been proposed for describing the connections and causal interactions between different regions of the brain. The goal of these models is to accurately mimic the event-related potentials observed by EEG/MEG measurement systems, and are useful in understanding overall brain functionality. In this paper, we focus on a class of nonlinear dynamic causal models (DCM) that are described by a set of connectivity parameters. In practice, the DCM parameters are inferred using data obtained by an EEG or MEG sensor array in response to a certain event or stimulus, and the resulting estimates are used to analyze the strength and direction of the causal interactions between different brain regions. The usefulness of the parameter estimates will depend on how accurately they can be estimated, which in turn will depend on noise, the sampling rate, number of data samples collected, the accuracy of the source localization and reconstruction steps, etc. The goal of this paper is to derive Cramér-Rao performance bounds for DCM estimates, and examine the behavior of the bounds under different operating conditions.

EEG/MEG source localization using source deflated matching pursuit.

A matching pursuit (MP) based algorithm, called source deflated matching pursuit (SDMP), is proposed for locating sources of brain activity. By iteratively deflating the contribution of identified sources to multiple measurement vectors (MMVs), the SDMP algorithm transforms the original multi-basis-vector/matrix selection problem into a single-basis-vector/matrix selection problem, which not only mitigates the residual-source interference but also remedies the intrinsic bias when locating deep sources. The robustness of the proposed algorithm to two bias factors is verified through simulations.

Matched subspace detector based feature extraction for sorting of multi-sensor action potentials.

This paper proposes a novel matched subspace detector (MSD) based algorithm for extracting discriminant features from multi-sensor measurements of extracellular action potentials (APs) to facilitate their subsequent separation according to the neuron of origin. The method does not require the construction of AP templates, and is therefore suitable for unsupervised AP sorting applications. In addition, detailed simulations show that the proposed algorithm outperforms existing single-sensor based feature extraction approaches.

Tracking single dynamic MEG dipole sources using the projected Extended Kalman Filter.

This paper presents two new algorithms based on the Extended Kalman Filter (EKF) for tracking the parameters of single dynamic magnetoencephalography (MEG) dipole sources. We assume a dynamic MEG dipole source with possibly both time-varying location and dipole orientation. The standard EKF-based tracking algorithm performs well under the assumption that the dipole source components vary in time as a Gauss-Markov process, provided that the background noise is temporally stationary. We propose a Projected-EKF algorithm that is adapted to a more forgiving condition where the background noise is temporally nonstationary, as well as a Projected-GLS-EKF algorithm that works even more universally, when the dipole components vary arbitrarily from one sample to the next.

Signal source localization with tetrodes: experimental verification.

Multi-sensor electrodes for extracellular recording of neuronal action potentials have significantly increased the signal-to-noise ratio (SNR) in neurophysiological experiments, ultimately leading to a more accurate interpretation of scientific data. Apart from improving SNR, we hypothesize that these electrodes can be used to estimate the location of underlying neuronal signal sources, and perhaps other parameters such as the size and shape of neurons whose activities are being recorded. This study introduces the multiple signal classification (MUSIC) algorithm to the problem of neuron localization and presents the first experimental demonstration of signal source localization using commercially available 4-sensor electrodes (tetrodes).