DualSort: online spike sorting with a running neural network.

Objective.Spike sorting, i.e. the detection and separation of measured action potentials from different extracellularly recorded neurons, remains one of the bottlenecks in deciphering the brain. In recent years, the application of neural networks (NNs) for spike sorting has garnered significant attention. Most methods focus on specific sub-problems within the conventional spike sorting pipeline, such as spike detection or feature extraction, and attempt to solve them with complex network architectures. This paper presents DualSort, a simple NN that gets combined with downstream post-processing for real-time spike sorting. It shows high efficiency, low complexity, and requires a comparatively small amount of human interaction.Approach.Synthetic and experimentally obtained extracellular single-channel recordings were utilized to train and evaluate the proposed NN. For training, spike waveforms were labeled with respect to their associated neuron and position in the signal, allowing the detection and categorization of spikes in unison. DualSort classifies a single spike multiple times in succession, as it runs over the signal in a step-by-step manner and uses a post-processing algorithm that transmits the network output into spike trains. Main results.With the used datasets, DualSort was able to detect and distinguish different spike waveforms and separate them from background activity. The post-processing algorithm significantly strengthened the overall performance of the model, making the system more robust as a whole. Although DualSort is an end-to-end solution that efficiently transforms filtered signals into spike trains, it competes with contemporary state-of-the-art technologies that exclusively target single sub-problems in the conventional spike sorting pipeline.Significance.This work demonstrates that even under high noise levels, complex NNs are not necessary by any means to achieve high performance in spike detection and sorting. The utilization of data augmentation on a limited quantity of spikes could substantially decrease hand-labeling compared to other studies. Furthermore, the proposed framework can be utilized without human interaction when combined with an unsupervised technique that provides pseudo labels for DualSort. Due to the low complexity of our network, it works efficiently and enables real-time processing on basic hardware. The proposed approach is not limited to spike sorting, as it may also be used to process different signals, such as electroencephalogram (EEG), which needs to be investigated in future research.

Estimation of Sparse VAR Models with Artificial Neural Networks for the Analysis of Biosignals.

Vector autoregressive models (VAR models) are often used to model and to analyze multivariate time series, especially to provide short-term forecasts. A common method of estimating coefficients of these VAR models is solving the Yule- Walker equations. This work introduces and investigates a method to set up "sparse" VAR models, in order to obtain a comparable prognosis quality with significantly fewer coefficients. For this purpose, an artificial neural network was programmed in Python with TensorFlow. Sparsity arises from the implementation of regularization algorithms.Based on simulated data and an ECG, we show that a comparable prognosis quality can be achieved with significantly fewer coefficients. In addition, sparse VAR models can also be determined if the data would actually lead to an underdetermined system of equations. Thus, sparse VAR models may help to classify short epochs of biosignals, e.g. P-waves or QRS-complexes.

[Physiological functional evaluation of retinal implants in animal models]. / Physiologische Funktionsprüfungen von Retinaimplantaten an Tiermodellen.

Retinal implants can--by electrical stimulation--create visual impressions in people with certain kinds of degenerative retinal diseases (e.g. Retinitis Pigmentosa). Electrically evoked potentials in the retina must be transferred into the visual cortex in an orderly manner, a prerequisite for any kind of form- and movement-perception. In the current developmental stage the difficult investigations are performed in various animal models: isolated retinae of intact chicken and of RCS-rats (a model for Retinitis Pigmentosa), as well as in anesthetised rabbits, pigs and cats with intact retinae. Our investigations show that spatially selective ganglion-cell responses can be recorded following focal electrical stimulation, in healthy and as well in degenerated retinae. Registration of activities in area 17 of the visual cortex demonstrate that electrical retinal stimulation can indeed activate it.

Implantation of retina stimulation electrodes and recording of electrical stimulation responses in the visual cortex of the cat.

BACKGROUND: Simple basic visual perception may be restored by epiretinal electrical stimulation in patients that are blind due to photoreceptor loss. To stimulate ganglion cells, epiretinally flat platinum microelectrodes embedded in thin polyimide film were developed and tested in the cat. METHODS: After removal of the lens and the vitreous body a thin microfilm electrode array was implanted through a corneoscleral incision in the cat eye (n = 4). In two eyes no further attempt was made to fixate the tip of the electrode, which was pressed onto the retinal surface due to the tension of the curved polyimide film. In two eyes the tip of the electrode was fixed with cyanoacrylate adhesive. The exterior part of the microelectrode film was directed under the skin towards the forehead which allowed fixation of the microplug to a head fixation bolt. Retinal stimulation experiments were performed within 1 week after implantation. Success of stimulation was assessed by recording neuronal activities from areas 17 and 18. Retinal microelectrodes were removed 2 weeks or longer after implantation. RESULTS: Intraocular inflammation or retinal detachment were not observed after implantation of the microelectrode film. In two eyes the tip of the microelectrodes dislocated spontaneously within the first few days. The lowest threshold of electrical stimulation was 35 microA, corresponding to a charge transfer of 14 nC per phase. These values were ten times higher than those obtained by needle electrodes used in prior experiments. CONCLUSIONS: Intraocular implanted flat microelectrodes made of platinum and polyimide were well tolerated. Because of the flat configuration of the microelectrodes higher stimulation thresholds than for needle electrodes were found, indicating insufficient contact to the retinal surface. An alternative shape and fixation technique is required to minimise electrodes' threshold of stimulation.

Phase correlation among rhythms present at different frequencies: spectral methods, application to microelectrode recordings from visual cortex and functional implications.

In classical EEG analysis rhythms with different frequencies occurring at separable regions and states of the brain are analysed. Rhythms in different frequency bands have often been assumed to be independent and their occurrence was interpreted as a sign of different functional operations. Independence has scarcely been proved because of conceptual and computational difficulties. It is, on the other hand, probable that different rhythmic brain processes are coupled because of the broad recurrent connectivity among brain structures. We, therefore, set out to find interactions among rhythmic signals at different frequencies. We were particularly interested in interactions between lower frequency bands and gamma-activities (30-90 Hz), because the latter have been analysed in our laboratory in great detail and had properties suggesting their involvement in perceptual feature linking. Fast oscillations occurred synchronized in a stimulus-specific way in the visual cortex of cat and monkey. Their presence was often accompanied by lower frequency components at considerable power. Such multiple spectral peaks are known from many cortical and subcortical structures. Despite their well known occurrence, coupling among different frequencies has not been established, apart from harmonic components. For the present investigation we extended existing analytical tools to detect non-linear correlations among signal pairs at any frequency (including incommensurate ones). These methods were applied to multiple microelectrode recordings from visual cortical areas 17 and 18 of anesthetized cats and V1 of awake monkeys. In particular, we assessed non-linear correlations by means of higher order spectral analysis of multi-unit spike activities (MUA) and local slow wave field potentials (LFP, 1-120 Hz) recorded with microelectrodes. Non-linear correlations among signal components at different frequencies were investigated in the following steps. First, the frequency content of short (approximately 250 ms) sliding window signal epochs was analyzed for simultaneously occurring rhythms of significant power at different frequencies. This was done by a newly developed method derived from the trispectrum using separate averaging of the products of short-epoch power spectra for any possible combination of frequency pairs. Second, non-linear (quadratic) phase coupling between different frequencies was assessed by the methods of bispectrum and bicoherence. We found phase correlations at different frequencies in the visual cortex of the cat and monkey. These couplings were significant in about 60% of the investigated MUA and LFP recordings, including several cases of coupling among incommensurate (i.e. non-harmonic) frequencies. Significant phase correlations were present: (1) within the gamma-frequency range; (2) between gamma- and low frequency ranges (1-30 Hz, including alpha- and beta-rhythms); and (3) within the low frequency range. Phase correlations depended, in most cases, on specific visual stimulation. We discuss the possible functional significance of phase correlations among high and low frequencies by including proposals from previous work about potential roles of single-frequency rhythms of the EEG. Our suggestions include: (1) visual feature linking across different temporal and spatial scales provided by coherent oscillations at high and low frequencies; (2) linking of visual cortical representations (high frequencies) to subcortical centers (low frequencies) like the thalamus and hippocampus; and (3) temporal segmentation of the sustained stream of incoming visual information into separate frames at different temporal resolutions in order to prevent perceptual smearing due to shifting retinal images. These proposals are, at present, merely speculative. However, they can, in principle, be proved by microelectrode recordings from trained behaving animals.