iRaster: A novel information visualization tool to explore spatiotemporal patterns in multiple spike trains

https://doi.org/10.1016/j.jneumeth.2010.09.009Get rights and content

Abstract

Over the last few years, simultaneous recordings of multiple spike trains have become widely used by neuroscientists. Therefore, it is important to develop new tools for analysing multiple spike trains in order to gain new insight into the function of neural systems. This paper describes how techniques from the field of visual analytics can be used to reveal specific patterns of neural activity. An interactive raster plot called iRaster has been developed. This software incorporates a selection of statistical procedures for visualization and flexible manipulations with multiple spike trains. For example, there are several procedures for the re-ordering of spike trains which can be used to unmask activity propagation, spiking synchronization, and many other important features of multiple spike train activity. Additionally, iRaster includes a rate representation of neural activity, a combined representation of rate and spikes, spike train removal and time interval removal. Furthermore, it provides multiple coordinated views, time and spike train zooming windows, a fisheye lens distortion, and dissemination facilities. iRaster is a user friendly, interactive, flexible tool which supports a broad range of visual representations. This tool has been successfully used to analyse both synthetic and experimentally recorded datasets. In this paper, the main features of iRaster are described and its performance and effectiveness are demonstrated using various types of data including experimental multi-electrode array recordings from the ganglion cell layer in mouse retina. iRaster is part of an ongoing research project called VISA (Visualization of Inter-Spike Associations) at the Visualization Lab in the University of Plymouth. The overall aim of the VISA project is to provide neuroscientists with the ability to freely explore and analyse their data. The software is freely available from the Visualization Lab website (see www.plymouth.ac.uk/infovis).

Research highlights

▶ Visual analytical improvements to the raster plot including: spike train reordering, overview, zoom, filter, and details-on-demand. ▶ A selection of statistical procedures for spike train reordering can reveal patterns. ▶ An augmented rate representation can reduce obfuscation in the raster plot. ▶ The analysis of multi-electrode recordings from the ganglion cell layer in mouse retina reveals a wave of propagating activity.

Introduction

A neuron, the fundamental element of the nervous system, is able to generate a short electrical impulse with a duration of 1–3 ms. The electrical impulse is called an action potential, or spike. A series of time moments t1, t2, …, tn, indicating consecutive impulses generated by the neuron is called a spike train. Multiple simultaneous spike trains can be recorded. For example, the Utah multi-electrode array (Jones et al., 1992) contains 100 microelectrodes and the recording from each electrode can contain information about the spikes of several neurons. Using a spike sorting method (e.g. Quiroga et al., 2004), it is possible to extract a set of several hundred simultaneous spike trains.

Spikes propagate across neural networks, forming neural representations of sensory information which are processed by the brain to underlie perception, movement, behaviour, and cognition (Robinson and Hall, 1998). Thus, the analysis of spike train data is crucial for understanding principles of information processing in the nervous system. An important goal of this analysis is to reveal any pattern of spiking activity which characterises a particular type of information processing. For example, a regime of synchronous spiking in the visual cortex correlates with attentional processing of visual information (Fries et al., 2001). A group of neurons representing features of the object in the attention focus demonstrates coherent spiking in some time interval. Therefore, a goal of the data analysis is to identify this subset of synchronously active neurons and to extract this data for further analysis. This problem is difficult because the pattern of coherent spiking is masked by the activities of other non-synchronous neurons. Also, the regime of synchronization is relatively short and it is difficult to scan through a whole epoch of a recording and find the relevant small time interval. Another example is a pattern of spiking corresponding to propagating activity. Let us suppose that some neuron, from the area of recording, receives an input and generates a spike (or a burst of spikes). This spike propagates to another neuron which also generates a spike that will, in turn, propagate to another neuron, etc. A pattern of propagating activity should be revealed but this pattern could be obscured by the activities of other recorded neurons. Our approach to solve these problems is based on the visualization of spiking information and particularly focused on ideas from the field of visual analytics.

A spike train can be regarded as a discrete time series, or point process on the time axis where points t1, t2, …, tn represent time moments of spike generation. Multiple spike trains are traditionally visualized by a raster plot (Awiszus, 1997). A raster plot is a two-dimensional representation of multiple spike trains. Examples of the raster plot are shown in Fig. 1.

The horizontal axis represents time and each spike train is visualized horizontally at the appropriate height on the vertical axis. Thus, along the vertical axis spike train identifiers are shown. Each individual spike (tik) is represented either by a dot or by a vertical bar, at its relevant location with coordinates (tik,yk), where (tik) is the time of ith spike in the kth spike train and (yk) is a vertical coordinate corresponding to the kth spike train. A raster plot only represents the event of spike generation, without showing the details of individual spike shape. In Fig. 1(top), a raster plot of 20 spike trains is shown by using the dot representation. The bottom raster plot of two spike trains is shown by using the bar representation. It is clear that the dot representation requires a smaller display region per spike train, however, the bar representation provides the information with greater clarity. Both of these raster plots approaches are commonly used.

We have developed a new tool called iRaster to help identify patterns of spiking activity. The software includes procedures for reordering spike trains and a selection of visual manipulations. Although iRaster includes various procedures based on known methods, the combination of them together with advanced and flexible visualization techniques result in unique novel software which supports the deciphering of various patterns of spiking activity from complex spike trains recorded from many simultaneous channels. In Section 2 we describe the state of the art and highlight the novelty of iRaster. In Section 3 we present the iRaster environment and in Section 3.1 spike train reordering procedures are described alongside a demonstration of how they work. Section 4 describes a selection of methods for the interactive use of iRaster. Section 5 presents the use of iRaster to analyse experimental data, resulting in the detection of a pattern of propagating activity. Finally, Section 6 concludes the paper.

Section snippets

Visualization of multiple spike trains using a raster plot

The raster plot is one of the core visualizations used by researchers to represent multidimensional spike train datasets. Classical raster plot visualizations are widely available in the most popular commercial applications used by researchers in this field. These tools include Matlab (The Mathworks, 2010) (with additional toolkits and libraries), Neuroexplorer (or Plexon) (Nex Technologies, 2010), pClamp (Molecular Devices, 2010) and STAToolkit (Weill Cornell Medical College, 2010). However,

The iRaster environment and detection of neural activity patterns

The main emphasis of this tool is the provision of flexibility to the user so that they can easily explore their increasingly large datasets. The main raster plot window supports all of the functionality of the information-seeking design mantra. It provides an overview of the entire dataset, it supports interactive zooming both horizontally and vertically (Fig. 2) and the user can also easily filter out data as required. Note that the user can exclude individual spike trains or consecutive

Highlight of important features of the iRaster

In this section, a selection of iRaster features are described. These features demonstrate the flexible and interactive visualization of multiple spike trains. For example, when analysing spiking activity it is useful to represent both spikes and activity rates at the same time. Thus, the augmented raster plot visualization is an important feature of iRaster.

Using iRaster to study experimental evidence of MEA recordings

In this section an example of how iRaster can be used to analyse experimental data is presented. An MEA comprising 59 electrodes was used to record spontaneous extracellular spiking activity. Spiking activity was recorded from the retina of a postnatal day 7 mouse in vitro. The electrodes were 30 μm in diameter with a gap of 200 μm between electrodes. The recording is 277 s in duration and Fig. 16 shows the corresponding raster plot of 59 spike trains. Each spike train is the recording of a single

Conclusions

This paper has presented a selection of techniques to help resolve the challenges of analysing spike train data within a raster plot. Reordering of spike trains is very significant to this type of analysis. This paper has shown that trends and anomalies in the data can be derived when the appropriate reordering procedures are applied to the data. In addition, the flexibility of the software has given absolute control to the user regarding the order of spike trains. This has enabled the user to

Acknowledgement

This work was partly supported by the EPSRC funded CARMEN project (EP/E002331/1).

References (20)

There are more references available in the full text version of this article.

Cited by (4)

View full text