Limits...
A model-based spike sorting algorithm for removing correlation artifacts in multi-neuron recordings.

Pillow JW, Shlens J, Chichilnisky EJ, Simoncelli EP - PLoS ONE (2013)

Bottom Line: Combining this measurement model with a Bernoulli prior over binary spike trains yields a posterior distribution for spikes given the recorded data.We introduce a greedy algorithm to maximize this posterior that we call "binary pursuit".The algorithm allows modest variability in spike waveforms and recovers spike times with higher precision than the voltage sampling rate.

View Article: PubMed Central - PubMed

Affiliation: Center for Perceptual Systems, Department of Psychology and Section of Neurobiology, The University of Texas at Austin, Austin, Texas, USA. pillow@mail.utexas.edu

ABSTRACT
We examine the problem of estimating the spike trains of multiple neurons from voltage traces recorded on one or more extracellular electrodes. Traditional spike-sorting methods rely on thresholding or clustering of recorded signals to identify spikes. While these methods can detect a large fraction of the spikes from a recording, they generally fail to identify synchronous or near-synchronous spikes: cases in which multiple spikes overlap. Here we investigate the geometry of failures in traditional sorting algorithms, and document the prevalence of such errors in multi-electrode recordings from primate retina. We then develop a method for multi-neuron spike sorting using a model that explicitly accounts for the superposition of spike waveforms. We model the recorded voltage traces as a linear combination of spike waveforms plus a stochastic background component of correlated Gaussian noise. Combining this measurement model with a Bernoulli prior over binary spike trains yields a posterior distribution for spikes given the recorded data. We introduce a greedy algorithm to maximize this posterior that we call "binary pursuit". The algorithm allows modest variability in spike waveforms and recovers spike times with higher precision than the voltage sampling rate. This method substantially corrects cross-correlation artifacts that arise with conventional methods, and substantially outperforms clustering methods on both real and simulated data. Finally, we develop diagnostic tools that can be used to assess errors in spike sorting in the absence of ground truth.

Show MeSH

Related in: MedlinePlus

Sensitivity of number of spikes recovered to the prior on spike rate.(A) Results for two example cells, one well-isolated (blue), and one poorly isolated (red). Adjusting the Bernoulli prior parameter (for each cell individually) alters the threshold used for spike identification (see Methods), which leads to an increase or decrease in the number of estimated spikes. (B) Simulation of detection of a scalar signal contaminated by Gaussian noise, for two different SNRs. Insets indicate histograms of noise observations (black) and signal observations (gray). The number of detections (“hits” plus “false positives”) varies with the choice of threshold, and the shape of the curve depends on the SNR. (C) Error rates (“misses” plus “false positives”) as a function of threshold for the simulations in (B).
© Copyright Policy
Related In: Results  -  Collection


getmorefigures.php?uid=PMC3643981&req=5

pone-0062123-g006: Sensitivity of number of spikes recovered to the prior on spike rate.(A) Results for two example cells, one well-isolated (blue), and one poorly isolated (red). Adjusting the Bernoulli prior parameter (for each cell individually) alters the threshold used for spike identification (see Methods), which leads to an increase or decrease in the number of estimated spikes. (B) Simulation of detection of a scalar signal contaminated by Gaussian noise, for two different SNRs. Insets indicate histograms of noise observations (black) and signal observations (gray). The number of detections (“hits” plus “false positives”) varies with the choice of threshold, and the shape of the curve depends on the SNR. (C) Error rates (“misses” plus “false positives”) as a function of threshold for the simulations in (B).

Mentions: We can use signal detection theory to develop a method for assessing the error rate of individual neurons in the absence of ground truth. This can be used both to select prior values for each cell, and to determine which neurons have acceptable spike sorting errors. The method is based on a simple observation: In a Bayesian setting, if an estimate is well constrained by the data, then the value of the prior parameter has little effect [45]. Thus, if the spike waveform of a cell is easily distinguished from the background noise and from the waveforms (or superpositions of waveforms) of other cells, the number of spikes found for that cell should be insensitive to the parameter value chosen. Fig. 6A illustrates this effect by showing the sensitivity of spike count to the Bernoulli prior parameter for two different RGCs. The well-isolated cell shows a spike count that is stable with respect to changes in threshold up to an order of magnitude in either direction. In contrast, the poorly-isolated cell is highly sensitive to the threshold value.


A model-based spike sorting algorithm for removing correlation artifacts in multi-neuron recordings.

Pillow JW, Shlens J, Chichilnisky EJ, Simoncelli EP - PLoS ONE (2013)

Sensitivity of number of spikes recovered to the prior on spike rate.(A) Results for two example cells, one well-isolated (blue), and one poorly isolated (red). Adjusting the Bernoulli prior parameter (for each cell individually) alters the threshold used for spike identification (see Methods), which leads to an increase or decrease in the number of estimated spikes. (B) Simulation of detection of a scalar signal contaminated by Gaussian noise, for two different SNRs. Insets indicate histograms of noise observations (black) and signal observations (gray). The number of detections (“hits” plus “false positives”) varies with the choice of threshold, and the shape of the curve depends on the SNR. (C) Error rates (“misses” plus “false positives”) as a function of threshold for the simulations in (B).
© Copyright Policy
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC3643981&req=5

pone-0062123-g006: Sensitivity of number of spikes recovered to the prior on spike rate.(A) Results for two example cells, one well-isolated (blue), and one poorly isolated (red). Adjusting the Bernoulli prior parameter (for each cell individually) alters the threshold used for spike identification (see Methods), which leads to an increase or decrease in the number of estimated spikes. (B) Simulation of detection of a scalar signal contaminated by Gaussian noise, for two different SNRs. Insets indicate histograms of noise observations (black) and signal observations (gray). The number of detections (“hits” plus “false positives”) varies with the choice of threshold, and the shape of the curve depends on the SNR. (C) Error rates (“misses” plus “false positives”) as a function of threshold for the simulations in (B).
Mentions: We can use signal detection theory to develop a method for assessing the error rate of individual neurons in the absence of ground truth. This can be used both to select prior values for each cell, and to determine which neurons have acceptable spike sorting errors. The method is based on a simple observation: In a Bayesian setting, if an estimate is well constrained by the data, then the value of the prior parameter has little effect [45]. Thus, if the spike waveform of a cell is easily distinguished from the background noise and from the waveforms (or superpositions of waveforms) of other cells, the number of spikes found for that cell should be insensitive to the parameter value chosen. Fig. 6A illustrates this effect by showing the sensitivity of spike count to the Bernoulli prior parameter for two different RGCs. The well-isolated cell shows a spike count that is stable with respect to changes in threshold up to an order of magnitude in either direction. In contrast, the poorly-isolated cell is highly sensitive to the threshold value.

Bottom Line: Combining this measurement model with a Bernoulli prior over binary spike trains yields a posterior distribution for spikes given the recorded data.We introduce a greedy algorithm to maximize this posterior that we call "binary pursuit".The algorithm allows modest variability in spike waveforms and recovers spike times with higher precision than the voltage sampling rate.

View Article: PubMed Central - PubMed

Affiliation: Center for Perceptual Systems, Department of Psychology and Section of Neurobiology, The University of Texas at Austin, Austin, Texas, USA. pillow@mail.utexas.edu

ABSTRACT
We examine the problem of estimating the spike trains of multiple neurons from voltage traces recorded on one or more extracellular electrodes. Traditional spike-sorting methods rely on thresholding or clustering of recorded signals to identify spikes. While these methods can detect a large fraction of the spikes from a recording, they generally fail to identify synchronous or near-synchronous spikes: cases in which multiple spikes overlap. Here we investigate the geometry of failures in traditional sorting algorithms, and document the prevalence of such errors in multi-electrode recordings from primate retina. We then develop a method for multi-neuron spike sorting using a model that explicitly accounts for the superposition of spike waveforms. We model the recorded voltage traces as a linear combination of spike waveforms plus a stochastic background component of correlated Gaussian noise. Combining this measurement model with a Bernoulli prior over binary spike trains yields a posterior distribution for spikes given the recorded data. We introduce a greedy algorithm to maximize this posterior that we call "binary pursuit". The algorithm allows modest variability in spike waveforms and recovers spike times with higher precision than the voltage sampling rate. This method substantially corrects cross-correlation artifacts that arise with conventional methods, and substantially outperforms clustering methods on both real and simulated data. Finally, we develop diagnostic tools that can be used to assess errors in spike sorting in the absence of ground truth.

Show MeSH
Related in: MedlinePlus