direkt zum Inhalt springen

direkt zum Hauptnavigationsmenü

Sie sind hier

TU Berlin

Page Content

Machine Learning

Selected Abstracts

An online algorithm for simultaneous spike detection and spike sorting based on matched filters and deconfusion
Citation key Natora2009
Author Natora, M. and Franke, F. and Daehne, S. and Obermayer, K.
Title of Book Goettingen Meeting of the German Neuroscience Society 2009
Year 2009
Abstract To understand higher cortical brain functions, an analysis of the simultaneous activity of a large number of individual neurons is essential. One common way to acquire the necessary amount of neural activity data is to use simultaneous extracellular recordings, either with single electrodes or, more recently, with multi electrode arrays. However, the recorded data does not directly provide the isolated activity of single neurons, but a mixture of neural activity from many neurons additionally corrupted with noise. The task of so called “spike sorting” algorithms is to reconstruct the single neuron signals (i.e. spike trains) from these recordings. In the past years many spike sorting algorithms have been developed. However there are only few algorithms which operate online and explicitly address the following needs: i) formulated for and making use of data from multi electrodes ii) high performance in detection and separation of overlapping spikes and iii) being able to adopt to non-stationarities of the data as caused by electrode drifts. We present a combined real-time online spike detection and spike sorting algorithm which explicitly address these three issues. A spike sorting algorithm for multi electrode data is formulated, which detects and resolves overlapping spikes with the same computational cost as non-overlapping spikes and tracks variations of the data. In an initial step, the number of neurons and their corresponding wave-form templates are estimated. From these templates a set of optimally matched filters is calculated. The filters constitute an approximation of an exact deconvolution (which is in general impossible for noisy data) and thus greatly improve the signalto- noise ratio. Instead of thresholding directly the output of each individual filter in order to detect/cluster spikes, we apply an additional transformation called deconfusion. In particular, by considering the maximal responses of all filters to every template, an un-mixing matrix is obtained. Similar to the ICA technique, this matrix is applied stepwise to the filter output providing a source separation. This minimizes the energy of the filter output of every filter to non-matching templates. Finally, a well-defined threshold is applied which leads to simultaneous spike detection and spike sorting. Our method needs an initialization phase, in which any supervised or unsupervised technique can be used to find the wave-form templates. Once the initial wave-forms are estimated, the algorithm can operate online and, because of the easy to implement linear operations, also in real-time. By incorporating a direct feedback from the newly found spikes to the existing templates, the matched filters are adapted constantly allowing for the detection of varying spikes shapes due to electrode drifts and non stationary noise characteristic. We evaluate the method on simulated and experimental data, including simultaneous extra/intra-cellular recordings and data from prefrontal cortex of behaving macaque monkeys. We compare the results to existing spike detection as well as spike sorting methods including thresholding combined with principal component analysis. We conclude that our algorithm is indeed able to resolve successfully overlapping spikes and outperforms other methods under realistic signal to noise ratios.
Download Bibtex entry

Zusatzinformationen / Extras

Quick Access:

Schnellnavigation zur Seite über Nummerneingabe

Auxiliary Functions