gms | German Medical Science

Objective: Neuronal prostheses are devices that can substitute parts of sensory, motor or cognitive modalities. High quality probing, recording and analysing the electrical signals in the brain will help to improve our understanding of neuronal information processing and thus will help to develop better neuroimplants. Here we report on a spike sorting software developed for multichannel extracellular fibre-electrode recordings.

Materials and Methods: We adopted and refined the spike-sorting algorithm by Franke et al.; 2010. For spike-detection we developed a multiresolution energy filter. Prior to feature extraction a correlation, Spainnergy based alignment of spikes is done. Feature extraction is realized via PCA on concatenated multichannel spikes. Clustering is done via a heteroscedastic Gaussian mixture model. To separate overlapping spikes we developed a novel deoverlapping algorithm. The software has been implemented in C++. For testing, simple models of neurons and electrodes and real recordings were used. Performance evaluation was done by using the fraction of correctly classified spikes.

Results: Performance is related to the number of electrodes, i.e., heptodes perform better than tetrodes for signals with many neurons or low signal-to-noise ratio. The correct number of PC is important for good classification. Deoverlapping enhances the performance, especially when neurons fire with high rates. The software is not yet ready for real-time spike sorting.

Discussion: We have developed algorithms and software for spike-sorting of multichannel extracellular recordings. In doing so, we solved the problem of deoverlapping superimposed spikes. Our results show, that our software can be used for off-line analysis. Future steps will focus on real-time spike sorting as required for neuronal prostheses.

Acknowledgment: Supported by BMWi ZIM grant KF2268909AK2.