hide
Free keywords:
-
Abstract:
Numerous approaches and methods have been introduced to solve the difficult task of sorting extracellular spike waveforms, which is a prerequisite for analyzing the contribution of individual cells to neuronal ensemble codes. However, the performance of many methods dramatically degrades under conditions of highly similar spike waveforms emitted from neighboring cells. An entirely unresolved problem is the separation of temporally overlapping action potentials which are quite numerous and constitute a very important feature of cortical processing. Most established sorting methods classify stable synchronous spikes as separate units and eliminate more variable near synchronous spikes as instable signal. What is missing is a quantitative evaluation of sorting performance by different algorithms for real data containing partially or completely overlapping spikes with reference to ground truth of individual spike timings. We set out to determine the performance of a sorting method based on independent component analysis (ICA) by quantitative analysis of spike times obtained from simultaneous tetrode and double patch clamp recordings in rat cortical slices. To this end, two pyramidal cells were patched and recorded simultaneously together with the extracellular signals of a tetrode placed in their vicinity under IR-DIC microscopic control. Action potentials were induced by current injection into both cells so that variable degrees of temporal overlap were generated. We used an ICA-based method because 1. our tetrode signals were sampled in a real 3D space (the 4 leads being arranged in a tetrahedron) 2. the ICA can fully exploit the spatial information of these tetrode signals. Separation performance on four data sets, each containing more than 3000 spikes of which 44.8%-57.6% (mean 52.6%) of spikes were overlapping, ranged from 87% to 97% (mean 92%). False negatives were on average observed for 0.4% of all events (due to noise in ICA decomposition data, making event detection difficult), false positives on average for 0.8% of identified events. The latter represent spuriously detected events in the ICA decomposition, which may be artifacts or spikes from other nearby cells. In sum, extracellular tetrode recordings of two neighboring cortical neurons can be sorted with a very high degree of confidence. In particular, this holds true for heavily overlapping spikes, producing indiscriminately looking waveform shapes in the extracellular tetrode signal. We conclude that the ICA model fits well on extracellular tetrode recording and enables separation of temporally overlapping action potentials from a limited number of neurons.
