Limits...
Automatic online spike sorting with singular value decomposition and fuzzy C-mean clustering.

Oliynyk A, Bonifazzi C, Montani F, Fadiga L - BMC Neurosci (2012)

Bottom Line: The results of these tests showed an excellent accuracy in discriminating low-amplitude and overlapping spikes under strong background noise.This new software provides neuroscience laboratories with a new tool for fast and robust online classification of single neuron activity.This feature could become crucial in situations when online spike detection from multiple electrodes is paramount, such as in human clinical recordings or in brain-computer interfaces.

View Article: PubMed Central - HTML - PubMed

Affiliation: Section of Human Physiology, Department of Biomedical Sciences and Advanced Therapies, Faculty of Medicine, University of Ferrara, Via Fossato di Mortara 17/19, 44121, Ferrara, Italy. lynnry@unife.it

ABSTRACT

Background: Understanding how neurons contribute to perception, motor functions and cognition requires the reliable detection of spiking activity of individual neurons during a number of different experimental conditions. An important problem in computational neuroscience is thus to develop algorithms to automatically detect and sort the spiking activity of individual neurons from extracellular recordings. While many algorithms for spike sorting exist, the problem of accurate and fast online sorting still remains a challenging issue.

Results: Here we present a novel software tool, called FSPS (Fuzzy SPike Sorting), which is designed to optimize: (i) fast and accurate detection, (ii) offline sorting and (iii) online classification of neuronal spikes with very limited or human intervention. The method is based on a combination of Singular Value Decomposition for fast and highly accurate pre-processing of spike shapes, unsupervised Fuzzy C-mean, high-resolution alignment of extracted spike waveforms, optimal selection of the number of features to retain, automatic identification the number of clusters, and quantitative quality assessment of resulting clusters independent on their size. After being trained on a short testing data stream, the method can reliably perform supervised online classification and monitoring of single neuron activity. The generalized procedure has been implemented in our FSPS spike sorting software (available free for non-commercial academic applications at the address: http://www.spikesorting.com) using LabVIEW (National Instruments, USA). We evaluated the performance of our algorithm both on benchmark simulated datasets with different levels of background noise and on real extracellular recordings from premotor cortex of Macaque monkeys. The results of these tests showed an excellent accuracy in discriminating low-amplitude and overlapping spikes under strong background noise. The performance of our method is competitive with respect to other robust spike sorting algorithms.

Conclusions: This new software provides neuroscience laboratories with a new tool for fast and robust online classification of single neuron activity. This feature could become crucial in situations when online spike detection from multiple electrodes is paramount, such as in human clinical recordings or in brain-computer interfaces.

Show MeSH

Related in: MedlinePlus

Distribution of-norm values of input singular vectors for every Class. This figure shows the frequency distribution of-norm values of singular vectors for each determined class in datasets A and B, respectively. Different classes are shown in different colours, according to the outcome of the clustering algorithm.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC3473300&req=5

Figure 9: Distribution of-norm values of input singular vectors for every Class. This figure shows the frequency distribution of-norm values of singular vectors for each determined class in datasets A and B, respectively. Different classes are shown in different colours, according to the outcome of the clustering algorithm.

Mentions: The implementation of full SVD (Eq.2) on the dataset A gave origin to matrices U(6191×24), S(24×24) and VT (6191×24). The elaboration of singular values in diagonal S matrix by algorithm determining the optimal number of features detected noc = 4 PCs accounting 67,3% of total variance. Then, the software calculates-norm values for every left singular vector (contains PCs) composing the matrix. The distribution of these-norm values showed four detectable peaks on the line-graph histogram (see Figure 9A), thus four clusters available at the PCs feature space (Figure 10A,B). The matrix VT and other parameters for the input of FCM clustering algorithm found on dataset A were kept in memory and then used to classify the spikes of dataset B. The Lratio measure of quality of clustering remained below the threshold value of 5 (Table 3) and so the software algorithm did not detect the need to repeat the testing phase of the clustering algorithm until the end of experiment B. The 3D scatter plot of first three PCs of dataset B shows same four clusters (Figure 10C). While two dense and partially overlapping clusters, located in the leftmost part of each plot, seem to be almost identical, the other two become more spread and shifted in 3D space in the dataset B with respect to dataset A, as it follows also from the Figure 11 representing the time course of clusters. We projected these clusters onto PCs axis of maximum variance and plotted this projection for each trial that were recorded sequentially. Figure 11B shows that a negative trend occurs in the projection, while Figure 11C shows the positive one. The solid and dash lines represents the centroid of the cluster as it changes over time for dataset A and B, respectively. The points in the figure represent a single spike waveform changing shape slowly and continuously. The changes in the internal structure of the dataset B become more evident considering the distribution of-norm values of singular vectors depicted in the line-histogram at the Figure 9B as Classes 1 and 4 become completely overlapped. Despite this fact all clusters are still well recognized and associated correctly by our approach thanks to cluster information previously learned in dataset A (Figure 10D).


Automatic online spike sorting with singular value decomposition and fuzzy C-mean clustering.

Oliynyk A, Bonifazzi C, Montani F, Fadiga L - BMC Neurosci (2012)

Distribution of-norm values of input singular vectors for every Class. This figure shows the frequency distribution of-norm values of singular vectors for each determined class in datasets A and B, respectively. Different classes are shown in different colours, according to the outcome of the clustering algorithm.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC3473300&req=5

Figure 9: Distribution of-norm values of input singular vectors for every Class. This figure shows the frequency distribution of-norm values of singular vectors for each determined class in datasets A and B, respectively. Different classes are shown in different colours, according to the outcome of the clustering algorithm.
Mentions: The implementation of full SVD (Eq.2) on the dataset A gave origin to matrices U(6191×24), S(24×24) and VT (6191×24). The elaboration of singular values in diagonal S matrix by algorithm determining the optimal number of features detected noc = 4 PCs accounting 67,3% of total variance. Then, the software calculates-norm values for every left singular vector (contains PCs) composing the matrix. The distribution of these-norm values showed four detectable peaks on the line-graph histogram (see Figure 9A), thus four clusters available at the PCs feature space (Figure 10A,B). The matrix VT and other parameters for the input of FCM clustering algorithm found on dataset A were kept in memory and then used to classify the spikes of dataset B. The Lratio measure of quality of clustering remained below the threshold value of 5 (Table 3) and so the software algorithm did not detect the need to repeat the testing phase of the clustering algorithm until the end of experiment B. The 3D scatter plot of first three PCs of dataset B shows same four clusters (Figure 10C). While two dense and partially overlapping clusters, located in the leftmost part of each plot, seem to be almost identical, the other two become more spread and shifted in 3D space in the dataset B with respect to dataset A, as it follows also from the Figure 11 representing the time course of clusters. We projected these clusters onto PCs axis of maximum variance and plotted this projection for each trial that were recorded sequentially. Figure 11B shows that a negative trend occurs in the projection, while Figure 11C shows the positive one. The solid and dash lines represents the centroid of the cluster as it changes over time for dataset A and B, respectively. The points in the figure represent a single spike waveform changing shape slowly and continuously. The changes in the internal structure of the dataset B become more evident considering the distribution of-norm values of singular vectors depicted in the line-histogram at the Figure 9B as Classes 1 and 4 become completely overlapped. Despite this fact all clusters are still well recognized and associated correctly by our approach thanks to cluster information previously learned in dataset A (Figure 10D).

Bottom Line: The results of these tests showed an excellent accuracy in discriminating low-amplitude and overlapping spikes under strong background noise.This new software provides neuroscience laboratories with a new tool for fast and robust online classification of single neuron activity.This feature could become crucial in situations when online spike detection from multiple electrodes is paramount, such as in human clinical recordings or in brain-computer interfaces.

View Article: PubMed Central - HTML - PubMed

Affiliation: Section of Human Physiology, Department of Biomedical Sciences and Advanced Therapies, Faculty of Medicine, University of Ferrara, Via Fossato di Mortara 17/19, 44121, Ferrara, Italy. lynnry@unife.it

ABSTRACT

Background: Understanding how neurons contribute to perception, motor functions and cognition requires the reliable detection of spiking activity of individual neurons during a number of different experimental conditions. An important problem in computational neuroscience is thus to develop algorithms to automatically detect and sort the spiking activity of individual neurons from extracellular recordings. While many algorithms for spike sorting exist, the problem of accurate and fast online sorting still remains a challenging issue.

Results: Here we present a novel software tool, called FSPS (Fuzzy SPike Sorting), which is designed to optimize: (i) fast and accurate detection, (ii) offline sorting and (iii) online classification of neuronal spikes with very limited or human intervention. The method is based on a combination of Singular Value Decomposition for fast and highly accurate pre-processing of spike shapes, unsupervised Fuzzy C-mean, high-resolution alignment of extracted spike waveforms, optimal selection of the number of features to retain, automatic identification the number of clusters, and quantitative quality assessment of resulting clusters independent on their size. After being trained on a short testing data stream, the method can reliably perform supervised online classification and monitoring of single neuron activity. The generalized procedure has been implemented in our FSPS spike sorting software (available free for non-commercial academic applications at the address: http://www.spikesorting.com) using LabVIEW (National Instruments, USA). We evaluated the performance of our algorithm both on benchmark simulated datasets with different levels of background noise and on real extracellular recordings from premotor cortex of Macaque monkeys. The results of these tests showed an excellent accuracy in discriminating low-amplitude and overlapping spikes under strong background noise. The performance of our method is competitive with respect to other robust spike sorting algorithms.

Conclusions: This new software provides neuroscience laboratories with a new tool for fast and robust online classification of single neuron activity. This feature could become crucial in situations when online spike detection from multiple electrodes is paramount, such as in human clinical recordings or in brain-computer interfaces.

Show MeSH
Related in: MedlinePlus