Spike Sorting

By Sarah Gibson, Jack W. Judy, and Dejan Marković

NOTE: This is an overview of the article, which appeared in the January 2012 issue of the IEEE Signal Processing Magazine.
Click here to read the entire article.

Introduction

Extracellular recording, the technique of inserting an electrode into the extra- cellular tissue of the brain to record the activity of individual neurons (“single-unit activity”), is a common experimental method used by neuroscientists to study how the brain works. In recent years, researchers have also demonstrated its potential use in medical technologies for the treatment of disorders such as paralysis, epilepsy. and memory loss. Although most of these applications require single-unit activity, these electrodes record the activity from multiple neurons surrounding the electrode. Spike sorting is the process of separating this signal into single-unit activity. A number of algorithms for this purpose have been published over the years, but there is still no universally accepted solution. This article presents an overview of the spike-sorting problem, its current solutions, and the challenges that remain. Because of the increasing demand for chronically implanted spike-sorting hardware, the article also discusses implementation considerations.

The ability of extracellular recording to provide researchers with neuron-level activity combined with its relatively low level of difficulty to perform (as compared to intracellular recording, for example) have led extracellular recording to become the dominant experimental technique in many studies.

And over the past decade, the technique of extracellular recording has received additional attention as researchers have begun to tap into its potential use in medical technologies for the treatment of disorders such as paralysis, epilepsy, and even cognitive and memory loss.

Whether the application is basic science research or medical technology, the signals from individual neurons (“single-unit activity”) are often of particular interest. In basic science, for example, the researcher may require knowledge of single-unit activity to study how a type of neuron responds to a specific stimulus. Similarly, most neural prosthetic technologies employ some sort of “decoding” algorithm – which may decode movement, intentions, or memories – that typically operates on signals from individual neurons. But because of the sizes of recording electrodes, the recorded signal is the sum of the signals from several (two to ten) neurons surrounding the electrode (“multiunit activity,” illustrated in Figure 1).

The electrical signal recorded from a microelectrode is the sum of the postsynaptic and action-potential activity of many neurons in the surrounding area.
Fig.1. The electrical signal recorded from a microelectrode is the sum of the postsynaptic and action-potential activity of many neurons in the surrounding area.

Click to enlarge

In such cases, spike sorting, the process of separating multiunit activity into groups of single- unit activity, is necessary. Beyond the functional role that spike sorting serves, spike sorting is important in providing the data reduction required of on-chip, multichannel processors. Wireless transmission of raw electrode data at the rate generated cannot be achieved under the strict power limits to which implantable electronics are subject. The alternative is to use bulky cables which restrict the mobility of the subject being monitored. On-chip spike sorting would overcome this problem.

To set a background for the discussion of spike sorting, the article presents some basic physiology of the neuron (Fig. 2), the action potential, and the electrical signals received by extra-cellular electrodes.

Diagram of a neuron.
Fig.2. Diagram of a neuron. Action potentials begin at the axon hillock and propagate down the axon. Depolarization of the axon terminal then triggers the release of neurotransmitters into the synaptic cleft, in turn depolarizing the postsynaptic cell.

Click to enlarge

Spike Sorting

To obtain (multi)unit activity, the extracellular data is first band- pass filtered to remove the LFP and high-frequency noise. To then obtain single- unit activity, spike sorting must be performed by sending this raw data into the signal-processing chain shown in Figure 3. The first steps are spike detection, the process of separating spikes from background noise, and alignment, the process of aligning all detected spikes to a common temporal point relative to the spike waveform. Once the spikes have been identified, spike sorting can take place.

The signal processing chain used to obtain single-unit activity.
Fig.3. The signal processing chain used to obtain single-unit activity.

Click to enlarge

Challenges in Spike Sorting

There are many unique characteristics of neural recording that make classification of neural data more difficult than for other types of data. One such characteristic is that there is almost always a lack of any sort of “ground truth” (a training period that uses known data to define cluster boundaries before the automatic classification period begins). In extracellular recording, however, experimenters typically must play a more passive role; they can only observe the neural activity, they cannot influence it. (Neural activity can be influenced by electrical stimulation, but usually not with single-cell precision.) Thus, there is no ground truth to be utilized in training the algorithms.

Other challenges to extracellular recording described in the article include non-gaussian noise, nonstationarity of data, and overlapping spikes. Work still remains in finding optimal automatic, real-time, efficient, and accurate spike-sorting algorithms that address all the remaining challenges, Finding such a solution to the spike-sorting problem would finally allow reliable spike sorting to be performed in implantable hardware. Performing spike sorting in hardware, simultaneously on many channels, would provide researchers with whole new experimental paradigms. For example, this would allow wireless transmission of data, thereby eliminating the need for cables. This would open the door for new types of experiments in which the activity of the brain is investigated as animals move freely in enriched (and possibly even natural) environments. It may also allow for recording from species that have never before been recorded, such as freely flying bats. Finally, implantable spike-sorting hardware would bring medical technologies for the treatment of disorders such as paralysis, epilepsy, and even cognitive and memory loss closer to a reality.

ABOUT THE AUTHORS

Sarah Gibson (sarah@ee.ucla.edu) received a B.S. degree in electrical and computer engineering from Baylor University in 2005 and an M.S. degree in electrical engineering in 2008 from the University of California, Los Angeles, where she is currently working toward a Ph.D. degree in electrical engineering. Her research interests are in systems and techniques for neural signal processing.

Jack W. Judy (jjudy@ee.ucla.edu) received the B.S.E.E. degree (summa cum laude) from the University of Minnesota, Minneapolis in 1989, and the M.S. and Ph.D. degrees in electrical engineering from the University of California, Berkeley in 1994 and 1996, respectively. He has been on the faculty of the Electrical Engineering Department at the University of California, Los Angeles (UCLA), since 1997, where he is currently an associate professor. At UCLA, he is the chair of the MEMS and nanotechnology major field of the Electrical Engineering Department and the director of the UCLA Neuroengineering Training Program.

Dejan Marković (dejan@ee.ucla.edu) received the Dipl.Ing. ́degree from the University of Belgrade, Serbia, in 1998 and the M.S. and Ph.D. degrees from the University of California, Berkeley, in 2000 and 2006, respectively, all in electrical engineering. In 2006, he joined the faculty of the Electrical Engineering Department at the University of California, Los Angeles as an assistant professor. His current research is focused on digital integrated circuits and DSP architectures for parallel data processing in future radio and healthcare systems, design with post-CMOS devices, design optimization methods, and CAD flows.