Optimal discrimination and classification of neuronal action potential waveforms from multiunit, multichannel recordings using software-based linear filters

Deep Learning-Based Template Matching Spike Classification for Extracellular Recordings

We propose a deep learning-based spike sorting method for extracellular recordings. For analysis of extracellular single unit activity, the process of detecting and classifying action potentials called “spike sorting” has become essential. This is achieved through distinguishing the morphological differences of the spikes from each neuron, which arises from the differences of the surrounding environment and characteristics of the neurons. However, cases of high structural similarity and noise make the task difficult. And for manual spike sorting, it requires professional knowledge along with extensive time cost and suffers from human bias. We propose a deep learning-based spike sorting method on extracellular recordings from a single electrode that is efficient, robust to noise, and accurate. In circumstances where labelled data does not exist, we created pseudo-labels through principal component analysis and K-means clustering to be used for multi-layer perceptron training and built high performing spike classification model. When tested, our model outperformed conventional methods by 2.1% on simulation data of various noise levels, by 6.0% on simulation data of various clusters count, and by 1.7% on in-vivo data. As a result, we showed that the deep learning-based classification can classify spikes from extracellular recordings, even showing high classification accuracy on spikes that are difficult even for manual classification.

Spike detection and spike sorting with a hidden Markov model improves offline decoding of motor cortical recordings

OBJECTIVE

Detection and sorting (classification) of action potentials from extracellular recordings are two important pre-processing steps for brain-computer interfaces (BCIs) and some neuroscientific studies. Traditional approaches perform these two steps serially, but using shapes of action potential waveforms during detection, i.e. combining the two steps, may lead to better performance, especially during high noise. We propose a hidden Markov model (HMM) based method for combined detecting and sorting of spikes, with the aim of improving the final decoding accuracy of BCIs.

APPROACH

The states of the HMM indicate whether there is a spike, what unit a spike belongs to, and the time course within a waveform. The HMM outputs probabilities of spike detection, and from this we can calculate expectations of spike counts in time bins, which can replace integer spike counts as input to BCI decoders. We evaluate the HMM method on simulated spiking data. We then examine the impact of using this method on decoding real neural data recorded from primary motor cortex of two Rhesus monkeys.

MAIN RESULTS

Our comparisons on simulated data to detection-then-sorting approaches and combined detection-and-sorting algorithms indicate that the HMM method performs more accurately at detection and sorting (0.93 versus 0.73 spike count correlation, 0.73 versus 0.49 adjusted mutual information). On real neural data, the HMM method led to higher adjusted mutual information between spike counts and kinematics (monkey K: 0.034 versus 0.027; monkey M: 0.033 versus 0.022) and better neuron encoding model predictions (K: 0.016 dB improvement; M: 0.056 dB improvement). Lastly, the HMM method facilitated higher offline decoding accuracy (Kalman filter, K: 8.5% mean squared error reduction, M: 18.6% reduction).

SIGNIFICANCE

The HMM spike detection and sorting method offers a new approach to spike pre-processing for BCIs and neuroscientific studies.

Neural Dynamics of Variable Grasp-Movement Preparation in the Macaque Frontoparietal Network

Our voluntary grasping actions lie on a continuum between immediate action and waiting for the right moment, depending on the context. Therefore, studying grasping requires an investigation into how preparation time affects this process. Two macaque monkeys (Macaca mulatta; one male, one female) performed a grasping task with a short instruction followed by an immediate or delayed go cue (0-1300 ms) while we recorded in parallel from neurons in the grasp preparation relevant area F5 that is part of the ventral premotor cortex, and the anterior intraparietal area (AIP). Initial population dynamics followed a fixed trajectory in the neural state space unique to each grip type, reflecting unavoidable movement selection, then diverged depending on the delay, reaching unique states not achieved for immediately cued movements. Population activity in the AIP was less dynamic, whereas F5 activity continued to evolve throughout the delay. Interestingly, neuronal populations from both areas allowed for a readout tracking subjective anticipation of the go cue that predicted single-trial reaction time. However, the prediction of reaction time was better from F5 activity. Intriguingly, activity during movement initiation clustered into two trajectory groups, corresponding to movements that were either "as fast as possible" or withheld movements, demonstrating a widespread state shift in the frontoparietal grasping network when movements must be withheld. Our results reveal how dissociation between immediate and delay-specific preparatory activity, as well as differentiation between cortical areas, is possible through population-level analysis.SIGNIFICANCE STATEMENT Sometimes when we move, we consciously plan our movements. At other times, we move instantly, seemingly with no planning at all. Yet, it's unclear how preparation for movements along this spectrum of planned and seemingly unplanned movement differs in the brain. Two macaque monkeys made reach-to-grasp movements after varying amounts of preparation time while we recorded from the premotor and parietal cortex. We found that the initial response to a grasp instruction was specific to the required movement, but not to the preparation time, reflecting required movement selection. However, when more preparation time was given, neural activity achieved unique states that likely related to withholding movements and anticipation of movement, shedding light on the roles of the premotor and parietal cortex in grasp planning.

The Impact of Anesthetic State on Spike-Sorting Success in the Cortex: A Comparison of Ketamine and Urethane Anesthesia

Spike sorting is an essential first step in most analyses of extracellular in vivo electrophysiological recordings. Here we show that spike-sorting success depends critically on characteristics of coordinated population activity that can differ between anesthetic states. In tetrode recordings from mouse auditory cortex, spike sorting was significantly less successful under ketamine/medetomidine (ket/med) than urethane anesthesia. Surprisingly, this difficulty with sorting under ket/med anesthesia did not appear to result from either greater millisecond-scale burstiness of neural activity or increased coordination of activity among neighboring neurons. Rather, the key factor affecting sorting success appeared to be the amount of coordinated population activity at long time intervals and across large cortical distances. We propose that spike-sorting success is directly dependent on overall coordination of activity, and is most disrupted by large-scale fluctuations in cortical population activity. Reliability of single-unit recording may therefore differ not only between urethane-anesthetized and ket/med-anesthetized states as demonstrated here, but also between synchronized and desynchronized states, asleep and awake states, or inattentive and attentive states in unanesthetized animals.

Combination of High-density Microelectrode Array and Patch Clamp Recordings to Enable Studies of Multisynaptic Integration

We present a novel, all-electric approach to record and to precisely control the activity of tens of individual presynaptic neurons. The method allows for parallel mapping of the efficacy of multiple synapses and of the resulting dynamics of postsynaptic neurons in a cortical culture. For the measurements, we combine an extracellular high-density microelectrode array, featuring 11’000 electrodes for extracellular recording and stimulation, with intracellular patch-clamp recording. We are able to identify the contributions of individual presynaptic neurons - including inhibitory and excitatory synaptic inputs - to postsynaptic potentials, which enables us to study dendritic integration. Since the electrical stimuli can be controlled at microsecond resolution, our method enables to evoke action potentials at tens of presynaptic cells in precisely orchestrated sequences of high reliability and minimum jitter. We demonstrate the potential of this method by evoking short- and long-term synaptic plasticity through manipulation of multiple synaptic inputs to a specific neuron.

Recent progress in multi-electrode spike sorting methods

In recent years, arrays of extracellular electrodes have been developed and manufactured to record simultaneously from hundreds of electrodes packed with a high density. These recordings should allow neuroscientists to reconstruct the individual activity of the neurons spiking in the vicinity of these electrodes, with the help of signal processing algorithms. Algorithms need to solve a source separation problem, also known as spike sorting. However, these new devices challenge the classical way to do spike sorting. Here we review different methods that have been developed to sort spikes from these large-scale recordings. We describe the common properties of these algorithms, as well as their main differences. Finally, we outline the issues that remain to be solved by future spike sorting algorithms.

Uniting functional network topology and oscillations in the fronto-parietal single unit network of behaving primates

The functional communication of neurons in cortical networks underlies higher cognitive processes. Yet, little is known about the organization of the single neuron network or its relationship to the synchronization processes that are essential for its formation. Here, we show that the functional single neuron network of three fronto-parietal areas during active behavior of macaque monkeys is highly complex. The network was closely connected (small-world) and consisted of functional modules spanning these areas. Surprisingly, the importance of different neurons to the network was highly heterogeneous with a small number of neurons contributing strongly to the network function (hubs), which were in turn strongly inter-connected (rich-club). Examination of the network synchronization revealed that the identified rich-club consisted of neurons that were synchronized in the beta or low frequency range, whereas other neurons were mostly non-oscillatory synchronized. Therefore, oscillatory synchrony may be a central communication mechanism for highly organized functional spiking networks.

DOI:http://dx.doi.org/10.7554/eLife.15719.001

The network of neurons in our brain generates all of our actions, yet it is not well understood how these neurons coordinate their activity with each other. Rhythmic electrical activity that happens at the same time across many different neurons is thought to be crucial for allowing different areas of the brain to communicate. However, it is still unclear what purpose rhythmic activity serves for communication. Are there groups of ‘hub’ neurons in different brain regions that coordinate overall activity by rhythmically synchronizing the network of neurons? Or is rhythmic activity insignificant for network coordination?

Dann et al. trained three monkeys to follow specific instructions to grasp a handle in different ways. While the monkeys performed the task, the activity of about 100 neurons was recorded simultaneously in three brain regions that are involved in planning and carrying out grasping movements. This revealed that the activity of the neurons was coordinated by a group of strongly connected hub neurons, which were distributed across all three of the brain regions. Nearly all of the hub neurons were rhythmically synchronized with each other, and also communicated with other neurons using rhythmic electrical activity.

Overall, the results presented by Dann et al. suggest that rhythmically synchronized activity is essential for neurons to coordinate how information is processed across the brain. Further studies into this method of communicating information will help to reveal how the primate brain can generate an immense range of behaviors.

DOI:http://dx.doi.org/10.7554/eLife.15719.002

A 16-Channel Nonparametric Spike Detection ASIC Based on EC-PC Decomposition

In extracellular neural recording experiments, detecting neural spikes is an important step for reliable information decoding. A successful implementation in integrated circuits can achieve substantial data volume reduction, potentially enabling a wireless operation and closed-loop system. In this paper, we report a 16-channel neural spike detection chip based on a customized spike detection method named as exponential component-polynomial component (EC-PC) algorithm. This algorithm features a reliable prediction of spikes by applying a probability threshold. The chip takes raw data as input and outputs three data streams simultaneously: field potentials, band-pass filtered neural data, and spiking probability maps. The algorithm parameters are on-chip configured automatically based on input data, which avoids manual parameter tuning. The chip has been tested with both in vivo experiments for functional verification and bench-top experiments for quantitative performance assessment. The system has a total power consumption of 1.36 mW and occupies an area of 6.71 mm (2) for 16 channels. When tested on synthesized datasets with spikes and noise segments extracted from in vivo preparations and scaled according to required precisions, the chip outperforms other detectors. A credit card sized prototype board is developed to provide power and data management through a USB port.

Predicting Reaction Time from the Neural State Space of the Premotor and Parietal Grasping Network

Neural networks of the brain involved in the planning and execution of grasping movements are not fully understood. The network formed by macaque anterior intraparietal area (AIP) and hand area (F5) of the ventral premotor cortex is implicated strongly in the generation of grasping movements. However, the differential role of each area in this frontoparietal network is unclear. We recorded spiking activity from many electrodes in parallel in AIP and F5 while three macaque monkeys (Macaca mulatta) performed a delayed grasping task. By analyzing neural population activity during action preparation, we found that state space analysis of simultaneously recorded units is significantly more predictive of subsequent reaction times (RTs) than traditional methods. Furthermore, because we observed a wide variety of individual unit characteristics, we developed the sign-corrected average rate (SCAR) method of neural population averaging. The SCAR method was able to explain at least as much variance in RT overall as state space methods. Overall, F5 activity predicted RT (18% variance explained) significantly better than AIP (6%). The SCAR methods provides a straightforward interpretation of population activity, although other state space methods could provide richer descriptions of population dynamics. Together, these results lend support to the differential role of the parietal and frontal cortices in preparation for grasping, suggesting that variability in preparatory activity in F5 has a more potent effect on trial-to-trial RT variability than AIP.

SIGNIFICANCE STATEMENT

Grasping movements are planned before they are executed, but how is the preparatory activity in a population of neurons related to the subsequent reaction time (RT)? A population analysis of the activity of many neurons recorded in parallel in macaque premotor (F5) and parietal (AIP) cortices during a delayed grasping task revealed that preparatory activity in F5 could explain a threefold larger fraction of variability in trial-to-trial RT than AIP. These striking differences lend additional support to a differential role of the parietal and premotor cortices in grasp movement preparation, suggesting that F5 has a more direct influence on trial-to-trial variability and movement timing, whereas AIP might be more closely linked to overall movement intentions.

Spike sorting of synchronous spikes from local neuron ensembles

Synchronous spike discharge of cortical neurons is thought to be a fingerprint of neuronal cooperativity. Because neighboring neurons are more densely connected to one another than neurons that are located further apart, near-synchronous spike discharge can be expected to be prevalent and it might provide an important basis for cortical computations. Using microelectrodes to record local groups of neurons does not allow for the reliable separation of synchronous spikes from different cells, because available spike sorting algorithms cannot correctly resolve the temporally overlapping waveforms. We show that high spike sorting performance of in vivo recordings, including overlapping spikes, can be achieved with a recently developed filter-based template matching procedure. Using tetrodes with a three-dimensional structure, we demonstrate with simulated data and ground truth in vitro data, obtained by dual intracellular recording of two neurons located next to a tetrode, that the spike sorting of synchronous spikes can be as successful as the spike sorting of nonoverlapping spikes and that the spatial information provided by multielectrodes greatly reduces the error rates. We apply the method to tetrode recordings from the prefrontal cortex of behaving primates, and we show that overlapping spikes can be identified and assigned to individual neurons to study synchronous activity in local groups of neurons.

A supervised multi-sensor matched filter for the detection of extracellular action potentials

Multi-sensor extracellular recording takes advantage of several electrode channels to record from multiple neurons at the same time. However, the resulting low signal-to-noise ratio (SNR) combined with biological noise makes signal detection, the first step of any neurophysiological data analysis, difficult. A matched filter was therefore designed to better detect extracellular action potentials (EAPs) from multi-sensor extracellular recordings. The detector was tested on tetrode data from a locust antennal lobe and assessed against three trained analysts. 25 EAPs and noise samples were selected manually from the data and used for training. To reduce complexity, the filter assumed that the underlying noise in the data was spatially white. The detector performed with an average TP and FP rate of 84.62% and 16.63% respectively. This high level of performance indicates the algorithm is suitable for widespread use.

On the robustness of EC-PC spike detection method for online neural recording

BACKGROUND

Online spike detection is an important step to compress neural data and perform real-time neural information decoding. An unsupervised, automatic, yet robust signal processing is strongly desired, thus it can support a wide range of applications. We have developed a novel spike detection algorithm called "exponential component-polynomial component" (EC-PC) spike detection.

NEW METHOD

We firstly evaluate the robustness of the EC-PC spike detector under different firing rates and SNRs. Secondly, we show that the detection Precision can be quantitatively derived without requiring additional user input parameters. We have realized the algorithm (including training) into a 0.13 μm CMOS chip, where an unsupervised, nonparametric operation has been demonstrated.

RESULTS

Both simulated data and real data are used to evaluate the method under different firing rates (FRs), SNRs. The results show that the EC-PC spike detector is the most robust in comparison with some popular detectors. Moreover, the EC-PC detector can track changes in the background noise due to the ability to re-estimate the neural data distribution.

COMPARISON WITH EXISTING METHODS

Both real and synthesized data have been used for testing the proposed algorithm in comparison with other methods, including the absolute thresholding detector (AT), median absolute deviation detector (MAD), nonlinear energy operator detector (NEO), and continuous wavelet detector (CWD). Comparative testing results reveals that the EP-PC detection algorithm performs better than the other algorithms regardless of recording conditions.

CONCLUSION

The EC-PC spike detector can be considered as an unsupervised and robust online spike detection. It is also suitable for hardware implementation.

Identification of a self-paced hitting task in freely moving rats based on adaptive spike detection from multi-unit M1 cortical signals

Invasive brain–computer interfaces (BCIs) may prove to be a useful rehabilitation tool for severely disabled patients. Although some systems have shown to work well in restricted laboratory settings, their usefulness must be tested in less controlled environments. Our objective was to investigate if a specific motor task could reliably be detected from multi-unit intra-cortical signals from freely moving animals. Four rats were trained to hit a retractable paddle (defined as a “hit”). Intra-cortical signals were obtained from electrodes placed in the primary motor cortex. First, the signal-to-noise ratio was increased by wavelet denoising. Action potentials were then detected using an adaptive threshold, counted in three consecutive time intervals and were used as features to classify either a “hit” or a “no-hit” (defined as an interval between two “hits”). We found that a “hit” could be detected with an accuracy of 75 ± 6% when wavelet denoising was applied whereas the accuracy dropped to 62 ± 5% without prior denoising. We compared our approach with the common daily practice in BCI that consists of using a fixed, manually selected threshold for spike detection without denoising. The results showed the feasibility of detecting a motor task in a less restricted environment than commonly applied within invasive BCI research.

High-density microelectrode array recordings and real-time spike sorting for closed-loop experiments: an emerging technology to study neural plasticity

Understanding plasticity of neural networks is a key to comprehending their development and function. A powerful technique to study neural plasticity includes recording and control of pre- and post-synaptic neural activity, e.g., by using simultaneous intracellular recording and stimulation of several neurons. Intracellular recording is, however, a demanding technique and has its limitations in that only a small number of neurons can be stimulated and recorded from at the same time. Extracellular techniques offer the possibility to simultaneously record from larger numbers of neurons with relative ease, at the expenses of increased efforts to sort out single neuronal activities from the recorded mixture, which is a time consuming and error prone step, referred to as spike sorting. In this mini-review, we describe recent technological developments in two separate fields, namely CMOS-based high-density microelectrode arrays, which also allow for extracellular stimulation of neurons, and real-time spike sorting. We argue that these techniques, when combined, will provide a powerful tool to study plasticity in neural networks consisting of several thousand neurons in vitro.

Applicability of independent component analysis on high-density microelectrode array recordings

Emerging complementary metal oxide semiconductor (CMOS)-based, high-density microelectrode array (HD-MEA) devices provide high spatial resolution at subcellular level and a large number of readout channels. These devices allow for simultaneous recording of extracellular activity of a large number of neurons with every neuron being detected by multiple electrodes. To analyze the recorded signals, spiking events have to be assigned to individual neurons, a process referred to as "spike sorting." For a set of observed signals, which constitute a linear mixture of a set of source signals, independent component (IC) analysis (ICA) can be used to demix blindly the data and extract the individual source signals. This technique offers great potential to alleviate the problem of spike sorting in HD-MEA recordings, as it represents an unsupervised method to separate the neuronal sources. The separated sources or ICs then constitute estimates of single-neuron signals, and threshold detection on the ICs yields the sorted spike times. However, it is unknown to what extent extracellular neuronal recordings meet the requirements of ICA. In this paper, we evaluate the applicability of ICA to spike sorting of HD-MEA recordings. The analysis of extracellular neuronal signals, recorded at high spatiotemporal resolution, reveals that the recorded data cannot be modeled as a purely linear mixture. As a consequence, ICA fails to separate completely the neuronal signals and cannot be used as a stand-alone method for spike sorting in HD-MEA recordings. We assessed the demixing performance of ICA using simulated data sets and found that the performance strongly depends on neuronal density and spike amplitude. Furthermore, we show how postprocessing techniques can be used to overcome the most severe limitations of ICA. In combination with these postprocessing techniques, ICA represents a viable method to facilitate rapid spike sorting of multidimensional neuronal recordings.

A computational model and simulation study of the efferent activity in the brachial nerves during voluntary motor intent

Inherent limitations of the surface myoelectric signal, such as the lack of recording sites in high-level amputations, and the sensitivity to placement and impedance effects, confound its wider application in powered prostheses. Since a functionally topographic distribution (somatotopic organization) of nerve fascicles exists within the peripheral nerves, it is theoretically possible that complete motor control information can be retrieved from peripheral nerve signals. In this study, we present a computational model that simulates the recording from specific nerve fascicles in the upper limb during voluntary contractions while they innervate relevant muscles. A procedure of classifying the nerve data is presented using a set of time domain features and a spike detection algorithm. Recommendations are made to achieve optimal neural signal recognition, with regard to electrode geometry and signal analysis.

Optimizing the automatic selection of spike detection thresholds using a multiple of the noise level

Thresholding is an often-used method of spike detection for implantable neural signal processors due to its computational simplicity. A means for automatically selecting the threshold is desirable, especially for high channel count data acquisition systems. Estimating the noise level and setting the threshold to a multiple of this level is a computationally simple means of automatically selecting a threshold. We present an analysis of this method as it is commonly applied to neural waveforms. Four different operators were used to estimate the noise level in neural waveforms and set thresholds for spike detection. An optimal multiplier was identified for each noise measure using a metric appropriate for a brain-machine interface application. The commonly used root-mean-square operator was found to be least advantageous for setting the threshold. Investigators using this form of automatic threshold selection or developing new unsupervised methods can benefit from the optimization framework presented here.

Robust unsupervised detection of action potentials with probabilistic models

We develop a robust and fully unsupervised algorithm for the detection of action potentials from extracellularly recorded data. Using the continuous wavelet transform allied to probabilistic mixture models and Bayesian probability theory, the detection of action potentials is posed as a model selection problem. Our technique provides a robust performance over a wide range of simulated conditions, and compares favorably to selected supervised and unsupervised detection techniques.

Automatic spike detection based on adaptive template matching for extracellular neural recordings

Recordings of extracellular neural activity are used in many clinical applications and scientific studies. In most cases, these signals are analyzed as a point process, and a spike detection algorithm is required to estimate the times at which action potentials occurred. Recordings from high-density microelectrode arrays (MEAs) and low-impedance microelectrodes often have a low signal-to-noise ratio (SNR<10) and contain action potentials from more than one neuron. We describe a new detection algorithm based on template matching that only requires the user to specify the minimum and maximum firing rates of the neurons. The algorithm iteratively estimates the morphology of the most prominent action potentials. It is able to achieve a sensitivity of >90% with a false positive rate of <5Hz in recordings with an estimated SNR=3, and it performs better than an optimal threshold detector in recordings with an estimated SNR>2.5.

Automated optimal detection and classification of neural action potentials in extra-cellular recordings

Determination of single unit spikes from multiunit spike trains plays a critical role in neurophysiological coding studies which require information about the precise timing of events underlying the neural codes that are the basis of behavior. Searching for optimal spike detection strategies has therefore been the focus of many studies over the past two decades. In this study we describe and implement an algorithm for the optimal real time detection and classification of neural spikes. The algorithm consists of three steps: noise analysis, template generation and real time detection and classification. The first step involves estimating the background noise statistics. In this step, a "cap-fitting" algorithm is used to automatically detect a spike free segment and then the mean, standard deviation and autocorrelation function of the noise are computed. The second step involves generating optimal templates of the spikes from a segment containing both noise and multiunit activity. In this step, a generalized matched filter is used to isolate a set of preliminary spikes from the noise. The first principal component of previously recorded templates is used as the deterministic signal. The preliminary spikes are then clustered in a sub-space spanned by the first three principal components to form new templates. The third step uses these templates for the real time spike detection and classification. In this step the incoming data are projected into a lower dimensional space that is designed to maximally separate the signal from the noise energy. This algorithm provides an accurate estimate of the signal to noise ratio and provides an accurate estimate of spike times and spike shapes.

Wavelet methods for spike detection in mouse renal sympathetic nerve activity

Abnormal autonomic nerve traffic has been associated with a number of peripheral neuropathies and cardiovascular disorders prompting the development of genetically altered mice to study the genetic and molecular components of these diseases. Autonomic function in mice can be assessed by directly recording sympathetic nerve activity. However, murine sympathetic spikes are typically detected using a manually adjusted voltage threshold and no unsupervised detection methods have been developed for the mouse. Therefore, we tested the performance of several unsupervised spike detection algorithms on simulated murine renal sympathetic nerve recordings, including an automated amplitude discriminator and wavelet-based detection methods which used both the discrete wavelet transform (DWT) and the stationary wavelet transform (SWT) and several wavelet threshold rules. The parameters of the wavelet methods were optimized by comparing basal sympathetic activity to postmortem recordings and recordings made during pharmacological suppression and enhancement of sympathetic activity. In general, SWT methods were found to outperform amplitude discriminators and DWT methods with similar wavelet coefficient thresholding algorithms when presented with simulations with varied mean spike rates and signal-to-noise ratios. A SWT method which estimates the noise level using a "noise-only" wavelet scale and then selectively thresholds scales containing the physiologically important signal information was found to have the most robust spike detection. The proposed noise-level estimation method was also successfully validated during pharmacological interventions.

A new action potential detector using the MTEO and its effects on spike sorting systems at low signal-to-noise ratios

This paper considers neural signal processing applied to extracellular recordings, in particular, unsupervised action potential detection at a low signal-to-noise ratio. It adopts the basic framework of the multiresolution Teager energy operator (MTEO) detector, but presents important new results including a significantly improved MTEO detector with some mathematical analyses, a new alignment technique with its effects on the whole spike sorting system, and a variety of experimental results. Specifically, the new MTEO detector employs smoothing windows normalized by noise power derived from mathematical analyses and has an improved complexity by utilizing the sampling rate. Experimental results prove that this detector achieves higher detection ratios at a fixed false alarm ratio than the TEO detector and the discrete wavelet transform detector. We also propose a method that improves the action potential alignment performance. Observing that the extreme points of the MTEO output are more robust to the background noise than those of the action potentials, we use the MTEO output for action potential alignment. This brings not only noticeable improvement in alignment performance but also quite favorable influence over the classification performance. Accordingly, the proposed detector improves the performance of the whole spike sorting system. We verified the improvement using various modeled neural signals and some real neural recordings.

Automated decomposition of intramuscular electromyographic signals

We present a novel method for extracting and classifying motor unit action potentials (MUAPs) from one-channel electromyographic recordings. The extraction of MUAP templates is carried out using a symbolic representation of waveforms, a common technique in signature verification applications. The assignment of MUAPs to their specific trains is achieved by means of repeated template matching passes using pseudocorrelation, a new matched-filter-based similarity measure. Identified MUAPs are peeled off and the residual signal is analyzed using shortened templates to facilitate the resolution of superimpositions. The program was tested with simulated data and with experimental signals obtained using fine-wire electrodes in the biceps brachii during isometric contractions ranging from 5% to 30% of the maximum voluntary contraction. Analyzed signals were made of up to 14 MUAP trains. Most templates were extracted automatically, but complex signals sometimes required the adjustment of 2 parameters to account for all the MUAP trains present. Classification accuracy rates for simulations ranged from an average of 96.3% +/- 0.9% (4 trains) to 75.6% +/- 11.0% (12 trains). The classification portion of the program never required user intervention. Decomposition of most 10-s-long signals required less than 10 s using a conventional desktop computer, thus showing capabilities for real-time applications.

Estimation of neural energy in microelectrode signals

We considered the problem of determining the neural contribution to the signal recorded by an intracortical electrode. We developed a linear least-squares approach to determine the energy fraction of a signal attributable to an arbitrary number of autocorrelation-defined signals buried in noise. Application of the method requires estimation of autocorrelation functions R(ap)(tau) characterizing the action potential (AP) waveforms and R(n)(tau) characterizing background noise. This method was applied to the analysis of chronically implanted microelectrode signals from motor cortex of rat. We found that neural (AP) energy consisted of a large-signal component which grows linearly with the number of threshold-detected neural events and a small-signal component unrelated to the count of threshold-detected AP signals. The addition of pseudorandom noise to electrode signals demonstrated the algorithm's effectiveness for a wide range of noise-to-signal energy ratios (0.08 to 39). We suggest, therefore, that the method could be of use in providing a measure of neural response in situations where clearly identified spike waveforms cannot be isolated, or in providing an additional 'background' measure of microelectrode neural activity to supplement the traditional AP spike count.

Optimal filtering for spike sorting of multi-site electrode recordings

We derive an optimal linear filter, to reduce the distortions of the peak amplitudes of action potentials in extracellular multitrode recordings, which are due to background activity and overlapping spikes. This filter is being learned very efficiently from the raw recordings in an unsupervised manner and responds to the average waveform with an impulse of minimal width. The average waveform does not have to be known in advance, but is learned together with the optimal filter. The peak amplitude of a filtered waveform is a more reliable estimate for the amplitude of an action potential than the peak of the biphasic waveform and can improve the accuracy of the event detection and clustering procedures. We demonstrate a spike-sorting application, in which events are detected using the Mahalanobis distance in the N-dimensional space of filtered recordings as a distance measure, and the event amplitudes of the filtered recordings are clustered to assign events to individual units. This method is fast and robust, and we show its performance by applying it to real tetrode recordings of spontaneous activity in the visual cortex of an anaesthetized cat and to realistic artificial data derived therefrom.

Evaluation of Spike-Detection Algorithms for a Brain-Machine Interface Application

Real time spike detection is an important requirement for developing brain machine interfaces (BMIs). We examined three classes of spike-detection algorithms to determine which is best suited for a wireless BMI with a limited transmission bandwidth and computational capabilities. The algorithms were analyzed by tabulating true and false detections when applied to a set of realistic artificial neural signals with known spike times and varying signal to noise ratios. A design-specific cost function was developed to score the relative merits of each detector; correct detections increased the score, while false detections and computational burden reduced it. Test signals both with and without overlapping action potentials were considered. We also investigated the utility of rejecting spikes that violate a minimum refractory period by occurring within a fixed time window after the preceding threshold crossing. Our results indicate that the cost-function scores for the absolute value operator were comparable to those for more elaborate nonlinear energy operator based detectors. The absolute value operator scores were enhanced when the refractory period check was used. Matched-filter-based detectors scored poorly due to their relatively large computational requirements that would be difficult to implement in a real-time system.

A wavelet-based method for action potential detection from extracellular neural signal recording with low signal-to-noise ratio

We present a method for the detection of action potentials, an essential first step in the analysis of extracellular neural signals. The low signal-to-noise ratio (SNR) and similarity of spectral characteristic between the target signal and background noise are obstacles to solving this problem and, thus, in previous studies on experimental neurophysiology, only action potentials with sufficiently large amplitude have been detected and analyzed. In order to lower the level of SNR required for successful detection, we propose an action potential detector based on a prudent combination of wavelet coefficients of multiple scales and demonstrate its performance for neural signal recording with varying degrees of similarity between signal and noise. The experimental data include recordings from the rat somatosensory cortex, the giant medial nerve of crayfish, and the cutaneous nerve of bullfrog. The proposed method was tested for various SNR values and degrees of spectral similarity. The method was superior to the Teager energy operator and even comparable to or better than the optimal linear detector. A detection ratio higher than 80% at a false alarm ratio lower than 10% was achieved, under an SNR of 2.35 for the rat cortex data where the spectral similarity was very high.

A micromachined silicon multielectrode for multiunit recording

A 16-channel multielectrode was used to record propagating action potentials from multiple units in the ventral nerve cord of the cricket Gryllus bimaculatus. The multielectrode was fabricated using photolithographic and bulk silicon etching techniques. The fabrication differs from standard methods in its use of deep reactive ion etching (DRIE) to form the bulk electrode structure. This technique enables the fabrication of relatively thick (>100 microm), rigid structures whose top surface can have any form of thin film electronics. The multielectrode tested in this paper consists of 16 narrow silicon bridges, 150 microm wide and 350 microm tall, spaced evenly over a centimeter, with passive rectangular gold recording sites on the top surface. The nerve cord was placed perpendicularly across the bridges. In this geometry, the nerve spans a 350 microm deep, 450 microm wide trench between each recording site, permitting adequate isolation of recording sites from each other and a platinum ground plane. Spike templates for eight neurons were formed using principle component analysis and clustering of the concatenated multichannel waveforms. Clean templates were generated from a 40 s recording of stimulus evoked activity. Conduction velocities ranged from 2.59+/-0.05 to 4.99+/-0.12 m/s. Two limitations of extracellular electrode arrays are the resolution of overlapping spikes and relation of discriminated units to known anatomy. The high density, precise positioning, and controlled impedance of recording sites achievable in microfabricated devices such as this one will aid in overcoming these limitations. The rigid devices fabricated using this process offer stable positioning of recording sites over relatively large distances (several millimeters) and are suitable for clamping or squeezing of nerve cords.

Neural spike sorting under nearly 0-dB signal-to-noise ratio using nonlinear energy operator and artificial neural-network classifier

We report a result on neural spike sorting under conditions where the signal-to-noise ratio is very low. The use of nonlinear energy operator enables the detection of an action potential, even when the SNR is so poor that a typical amplitude thresholding method cannot be applied. The superior detection ability facilitates the collection of a training set under lower SNR than that of the methods which employ simple amplitude thresholding. Thus, the statistical characteristics of the input vectors can be better represented in the neural-network classifier. The trained neural-network classifiers yield the correct classification ratio higher than 90% when the SNR is as low as 1.2 (0.8 dB) when applied to data obtained from extracellular recording from Aplysia abdominal ganglia using a semiconductor microelectrode array.

A 100-channel system for real time detection and storage of extracellular spike waveforms

As extracellular electrode arrays with 100 or more active recording sites become more widely used for simultaneous recording of neural ensembles, practical data acquisition systems that can efficiently accommodate high electrode counts are needed. To reduce the high data rates associated with extracellular recordings from these arrays, various algorithms and systems have been designed to provide complete online detection and classification of extracellular spike waveforms. However, many of these algorithms require significant user supervision to ensure accurate performance. In this paper, we discuss the design and validation of a 100-channel PC-based system that can be used with arrays of extracellular electrodes such as the Utah Electrode Array. Instead of comprehensive online spike analysis, the system performs online detection and storage of the spike waveforms for offline classification. This strategy preserves the data of interest, reduces system complexity, and requires less user supervision during experiments.

Protocol for microneurography with concentric needle electrodes

In 1968, the method of human percutaneous microneurography with solid tungsten electrodes was introduced. Since then many investigators used this technique to study peripheral mechanisms in the somatosensory, motor and autonomic systems of conscious humans. Although some modifications of the method were described, the basic construction of the recording electrode has remained the same over the years. In the present protocol we describe in detail the procedures of microneurography using a thin diameter concentric needle electrode. There are some advantages with the concentric electrodes in comparison with the tungsten needles: (1) the electrical and mechanical properties of the electrode are stable which allows repeated use, (2) its restricted and one-dimensionally directed recording area provides the possibility to study topographical aspects within even a part of a peripheral nerve fascicle, and (3) multi-channel recordings can be achieved by adding more recording surfaces to the electrode. Based on recent investigations evaluating the recording properties of concentric electrodes we propose a novel procedure for signal analysis where template matching is incorporated. The analyses described in this protocol might also be applicable for extracellular recordings from muscle or elsewhere within the nervous system.

In vitro 2-D networks of neurons characterized by processing the signals recorded with a planar microtransducer array

The purpose of this paper is to extensively analyze and utilize the key features that characterize the recently available electrophysiological technique of growing selected populations of neurons on planar substrate microelectrode arrays. This experimental configuration is first simulated by modeling the signal transduction operated by an array of microtransducers coupled to a network of Hodgkin-Huxley-like neurons, connected to each other with given levels of synaptic strength. Signal processing tools are then described and validated by identifying the various degrees of connectivity previously introduced into the simulated network. Finally, these software tools are utilized to characterize the activity and identify the synaptic connectivity of networks of cultured neurons extracted from dorsal root ganglia (DRG) of chick embryos and exposed to synapse inhibiting/reinforcing ions. As a result, correlations between various regimens of electrophysiological activity and synaptic strength are obtained.

A microcomputer program for automated neuronal spike detection and analysis

A system for on-line spike detection and analysis based on an IBM PC/AT compatible computer, written in TURBO PASCAL 6.0 and using commercially available analog-to-digital hardware is described here. Spikes are detected by an adaptive threshold which varies as a function of signal mean and its variability. Since the threshold value is determined automatically by the signal-to-noise ratio analysis, the user is not actively involved in controlling its level. This program has been reliably used for the detection and analysis of the spike discharge of vestibular system afferent neurons. It generates the interval-joint distribution graph, the interval histogram, the autocorrelation function, the autocorrelation histogram, and phase-space graphs, thus, providing a complete set of graphical and statistical data for the characterization of the dynamics of neuronal spike activity. Data can be exported to other software such as Excel, Sigmaplot and MatLab, for example.

Detection, classification, and superposition resolution of action potentials in multiunit single-channel recordings by an on-line real-time neural network

Determination of single-unit spike trains from multiunit recordings obtained during extracellular recording has been the focus of many studies over the last two decades. In multiunit recordings, superpositions can occur with high frequency if the firing rates of the neurons are high or correlated, making superposition resolution imperative for accurate spike train determination. In this work, a connectionist neural network (NN) was applied to the spike sorting challenge. A novel training scheme was developed which enabled the NN to resolve some superpositions using single-channel recordings. Simulated multiunit spike trains were constructed from templates and noise segments that were extracted from real extracellular recordings. The simulations were used to determine the performances of the NN and a simple matched template filter (MTF), which was used as a basis for comparison. The network performed as well as the MTF in identifying nonoverlapping spikes, and was significantly better in resolving superpositions and rejecting noise. An on-line, real-time implementation of the NN discriminator, using a high-speed digital signal processor mounted inside an IBM-PC, is now in use in six laboratories.

Correspondence of escape-turning behavior with activity of descending mechanosensory interneurons in the cockroach, Periplaneta americana

Two bilaterally paired mechanosensory neurons that respond to antennal touch stimulation recently have been described in the cockroach Periplaneta americana. Here chronic recordings were used to describe the activity of these interneurons in relation to behavior. Parallel intra/extracellular recording experiments showed that both pairs of previously identified descending mechanosensory interneurons (DMIs) were activated after touch stimulation of the antennae and before initiation of escape. On a trial-by-trial basis, the bilateral pattern of their activity was correlated with sensory input and behavior: when one antenna was touched, the contralateral DMI axons displayed impulses earlier and in greater numbers than their ipsilateral homologs; turns were made toward the side with greater DMI activity, i.e., away from the touched antenna. One parameter of DMI activity (the bilateral difference in number of DMI impulses) was correlated with the angular amplitude of turning. In the absence of touch stimulation, unilateral electrical stimulation of a cervical connective via the chronic electrodes produced turning movements similar to natural escape turning and of appropriate directionality. These results support the hypothesis that neural activity in DMIs is involved in the control of antennal touch-evoked escape, and they provide a basis for a model of DMI specification of the direction of escape turning.