Deep brain stimulation (DBS) is widely used as a treatment option for patients with movement disorders. In addition to its clinical impact, DBS has been utilized in the field of cognitive neuroscience, wherein the answers to several fundamental questions underpinning the mechanisms of neuromodulation in decision making rely on the ways in which a burst of DBS pulses, usually delivered at a clinical frequency, i.e., 130 Hz, perturb participants’ choices. It was observed that neural activities recorded during DBS were contaminated with large artifacts, which lasts for a few milliseconds, as well as a low-frequency (slow) signal (~1–2 Hz) that can persist for hundreds of milliseconds. While the focus of most of methods for removing DBS artifacts was on the former, the artifact removal capabilities of the slow signal have not been addressed. In this work, we propose a new method based on combining singular value decomposition (SVD) and normalized adaptive filtering to remove both large (fast) and slow artifacts in local field potentials, recorded during a cognitive task in which bursts of DBS were utilized. Using synthetic data, we show that our proposed algorithm outperforms four commonly used techniques in the literature, namely, (1) normalized least mean square adaptive filtering, (2) optimal FIR Wiener filtering, (3) Gaussian model matching, and (4) moving average. The algorithm’s capabilities are further demonstrated by its ability to effectively remove DBS artifacts in local field potentials recorded from the subthalamic nucleus during a verbal Stroop task, highlighting its utility in real-world applications.
The amount of information that differentially correlated spikes in a neural ensemble carry is not the same; the information of different types of spikes is associated with different features of the stimulus. By calculating a neural ensemble’s information in response to a mixed stimulus comprising slow and fast signals, we show that the entropy of synchronous and asynchronous spikes are different, and their probability distributions are distinctively separable. We further show that these spikes carry a different amount of information. We propose a time-varying entropy (TVE) measure to track the dynamics of a neural code in an ensemble of neurons at each time bin. By applying the TVE to a multiplexed code, we show that synchronous and asynchronous spikes carry information in different time scales. Finally, a decoder based on the Kalman filtering approach is developed to reconstruct the stimulus from the spikes. We demonstrate that slow and fast features of the stimulus can be entirely reconstructed when this decoder is applied to asynchronous and synchronous spikes, respectively. The significance of this work is that the TVE can identify different types of information (for example, corresponding to synchronous and asynchronous spikes) that might simultaneously exist in a neural code.
There is an increasing demand for a computationally efficient and accurate point process filter solution for real-time decoding of population spiking activity in multidimensional spaces. Real-time tools for neural data analysis, specifically real-time neural decoding solutions open doors for developing experiments in a closed-loop setting and more versatile brain-machine interfaces. Over the past decade, the point process filter has been successfully applied in the decoding of behavioral and biological signals using spiking activity of an ensemble of cells; however, the filter solution is computationally expensive in multi-dimensional filtering problems. Here, we propose an approximate filter solution for a general point-process filter problem when the conditional intensity of a cell's spiking activity is characterized using a Mixture of Gaussians. We propose the filter solution for a broader class of point process observation called marked point-process, which encompasses both clustered —mainly, called sorted— and clusterless —generally called unsorted or raw— spiking activity. We assume that the posterior distribution on each filtering time-step can be approximated using a Gaussian Mixture Model and propose a computationally efficient algorithm to estimate the optimal number of mixture components and their corresponding weights, mean, and covariance estimates. This algorithm provides a real-time solution for multi-dimensional point-process filter problem and attains accuracy comparable to the exact solution. Our solution takes advantage of mixture dropping and merging algorithms, which collectively control the growth of mixture components on each filtering time-step. We apply this methodology in decoding a rat's position in both 1-D and 2-D spaces using clusterless spiking data of an ensemble of rat hippocampus place cells. The approximate solution in 1-D and 2-D decoding is more than 20 and 4,000 times faster than the exact solution, while their accuracy in decoding a rat position only drops by less than 9% and 4% in RMSE and 95% HPD coverage performance metrics. Though the marked-point filter solution is better suited for real-time decoding problems, we discuss how the filter solution can be applied to sorted spike data to better reflect the proposed methodology versatility.
Recent technological and experimental advances in recording from neural systems have led to a significant increase in the type and volume of data being collected in neuroscience experiments. This brings an increasing demand for development of appropriate analytical tools to analyze large scale neuroscience data. Simultaneously, advancement in deep neural networks (DNNs) and statistical modeling frameworks have provided new techniques for analysis of diverse forms of neuroscience data. DNNs like Long short-term memory (LSTM) or statistical modeling approaches like state-space point-process (SSPP) are widely used in the analysis of neural data including neural coding and inference analysis. Despite wide utilization of these techniques, there is a lack of comprehensive studies which systematically assess attributes of LSTM and SSPP approaches on a common neuroscience data analysis problem. As a result, this occasionally leads to inconsistent and divergent conclusions on the strength or weakness of either of the methodologies and also statistical significance of the analytical outcomes. In this research, we focus on providing a more systematic and multifaceted assessment of LSTM and SSPP techniques in a neural decoding problem. We examine different settings and modeling specifications to attain the optimal modeling solutions. We propose new LSTM network topologies and approximate filter solution to estimate a rat movement trajectory in a 2-D spaces using an ensemble of place cells’ spiking activity. For each technique; we then study performance, computational efficiency, and generalizability of each technique in this decoding problem. By utilizing these results, we provided a succinct picture of the strength and weakness of each modeling approach and suggest who each of these techniques can be properly utilized in neural decoding problems.
The Bayesian state-space neural encoder-decoder modeling framework is an established solution to reveal how changes in brain dynamics encode physiological covariates like movement or cognition. Although the framework is increasingly being applied to progress the field of neuroscience, its application to modeling high-dimensional neural data continues to be a challenge. Here, we propose a novel solution that avoids the complexity of encoder models that characterize high-dimensional data as a function of the underlying state processes. We build a discriminative model to estimate state processes as a function of current and previous observations of neural activity. We then develop the filter and parameter estimation solutions for this new class of state-space modeling framework called the “direct decoder” model. We applied the model to decode movement trajectories of a rat in a W-shaped maze from the ensemble spiking activity of place cells and achieve comparable performance to modern decoding solutions, without needing an encoding step in the model development. We further demonstrate how a dynamical auto-encoder can be built using the direct decoder model; where the underlying state process links the high-dimensional neural activity to the behavioral readout. We applied the dynamical auto-encoder model in estimating the intention to verbally communicate of an epileptic participant and their companions. The result shows that the dynamical auto-encoder can optimally estimate the low-dimensional dynamical manifold which represents the relationship between the brain and behavior.