Bio-inspired benchmark generator for extracellular multi-unit recordings

Abstract

The analysis of multi-unit extracellular recordings of brain activity has led to the development of numerous tools, ranging from signal processing algorithms to electronic devices and applications. Currently, the evaluation and optimisation of these tools are hampered by the lack of ground-truth databases of neural signals. These databases must be parameterisable, easy to generate and bioinspired, i.e. containing features encountered in real electrophysiological recording sessions. Towards that end, this article introduces an original computational approach to create fully annotated and parameterised benchmark datasets, generated from the summation of three components: neural signals from compartmental models and recorded extracellular spikes, non-stationary slow oscillations, and a variety of different types of artefacts. We present three application examples. Electrical recording of extracellular action potentials is the "gold standard" technique widely used in electrophysiology 1 , where the signals are exploited to correlate neural activity with a behavioural output and/or the electrophysiological consequences of brain lesions or drug infusion, etc. The emergence of novel methods for neural analysis together with high-throughput data acquisition technologies 2 provide new possibilities for the exploitation of brain activity at the single unit level, for example, giving instantaneous feedback for closed-loop interactions with brain circuits when abnormal neural signals are detected 3 . This approach has proven effective for several pathological conditions such as Epilepsy, Parkinson's disease, or Essential Tremor 4-7 . From a more fundamental perspective, novel algorithms have been recently proposed to process these large amounts of neural data, such as semi-automatic and automatic clustering techniques, to distinguish different neural sources in multi-unit extracellular recordings [8][9][10][11][12] . In order to validate the performance and accuracy of these different algorithms or devices, reliable datasets, where the majority of the signal content is known, are essential. Ideally, this ground-truth reference should be a completely annotated and parameterised dataset, in which three levels of information should be modifiable and known in detail: the recording environment (e.g. density of active population of neurons or distance from neurons to recording sites), the population dynamics (e.g. firing rate, spike timing of each neuron and spike waveforms) and the noise content (e.g. background noise level contribution and number of artefacts).There are several applications (Fig. 1) where using a parameterised dataset can be advantageous, ranging from algorithm design to development and evaluation of electronic devices. Moreover, parameterised datasets are needed to evaluate the efficiency of unsupervised classification algorithms. In recent years, several spike sorting algorithms have been proposed [8][9][10][11][12] , however, it is difficult to assess their sorting efficiency since the datasets used to evalu...

MEArec: a fast and customizable testbench simulator for ground-truth extracellular spiking activity

“…A third validation method consist of using simulated ground-truth recordings [12]. While this approach is promising, in combination with experimental paired recordings, the current available simulators [7,19,35] present some limitations in terms of biological realism, controllability, speed, and/or ease of use (see Introduction). We therefore introduced MEArec, a software package which is computationally efficient, easy to use, highly controllable, and capable of reproducing critical characteristics of extracellular recordings relevant to spike sorting, including bursting modulation, spatio-temporal overlaps, drift of units over time, and various noise profiles.…”

“…In the last years, there have been a few projects aim to develop neural simulators for benchmarking spike sorting methods [7,19,35]: Camunas et al developed NeuroCube [7], a MATLAB-based simulator which combines biophysically detailed cell models and synthetic spike trains to simulate the activity of neurons close to a recording probe, while noise is simulated by the activity of distant neurons. NeuroCube is very easy to use with a simple and intuitive graphical user interface (GUI).…”

“…As the running of NEURON simulations with biophysically detailed neurons can be computationally expensive, the use of ViSAPy to generate long-duration spike-sorting benchmarking data is boosted by access to powerful computers. Mondragon et al developed a Neural Benchmark Simulator (NBS) [35] extending the NeuroCube software. NBS extends the capability of NeuroCube for using user-specific probes, and it combines the spiking activity signals (from NeuroCube), with low-frequency activity signals, and artifacts libraries shipped with the code.…”

MEArec: A Fast and Customizable Testbench Simulator for Ground-truth Extracellular Spiking Activity

“…In the last years, there have been a few projects aiming to develop neural simulators for benchmarking spike sorting methods (Camuñas-Mesa and Quiroga 2013 ; Hagen et al 2015 ; Mondragón-González and Burguière 2017 ): Camunas et al developed NeuroCube (Camuñas-Mesa and Quiroga 2013 ), a MATLAB-based simulator which combines biophysically detailed cell models and synthetic spike trains to simulate the activity of neurons close to a recording probe, while noise is simulated by the activity of distant neurons. NeuroCube is very easy to use with a simple and intuitive graphical user interface (GUI).…”

“…Mondragon et al developed a Neural Benchmark Simulator (NBS) (Mondragón-González and Burguière 2017 ) extending the NeuroCube software. NBS extends the capability of NeuroCube for using user-specific probes, and it combines the spiking activity signals (from NeuroCube), with low-frequency activity signals, and artifacts libraries shipped with the code.…”