Research ReportDouble sliding-window technique: A new method to calculate the neuronal response onset latency
Introduction
Peristimulus time histograms (PSTHs; developed by Gerstein and Kiang, 1960) are commonly used to visualize the effect of a stimulus on the neuronal activity in extracellular recordings. Estimation of the neuronal response onset latency may provide important data concerning the information flow within the central nervous system (Berson, 1987, Dreher and Sefton, 1979). Despite the fact that the response onset latency comprises a source of data with which to resolve the information coding in the central nervous system alternative to the well-discussed properties of neuronal responses, i.e. the neuronal firing frequency, response duration and stimulus threshold, only a small proportion of neuronal recordings are generally analyzed from this aspect, possibly because of the weakness of automated latency estimation methods.
In the automation of the estimation of latency, a basic problem is to extract a signal from the spontaneous activity, which is determined by environmental and physiological noise. Mathematically, two solutions exist: counting the number of impulses discharged in some fixed interval (counting method) and detecting the time for the discharge of a fixed number of impulses (timing method) (Wandell, 1977). Poisson spike-train analysis is currently the most frequently applied method of latency estimation (Legéndy and Salcman, 1985). The neuronal response onset is calculated by averaging the time positions of some arbitrarily chosen bursts in the proven trials. The cumulative sum (CUSUM) technique was the first method in which the latency of neuronal responses was calculated via the analysis of PSTHs (Ellaway, 1978). The value of this method lies in the detection of change in the mean level of the activity. Since the change in the mean could be small relative to the variation in the individual values, an arbitrary threshold level (usually 1, 2 or 3 standard deviations (SD) above the mean of the spontaneous activity condition) is often chosen to quantify the start-point of the increment in the CUSUM curve (Ouellette and Casanova, 2006). The CUSUM technique has the weakness that it is not possible to determine precisely which temporal component of the response should be analyzed. Accordingly, Falzett et al. (1985) introduced a combination of the CUSUM technique with a second-order difference (SOD) function. However, despite the numerous methods proposed (Table 1), none of them can be applied reassuringly as a universal latency estimation method.
Is it possible to develop an estimation method for onset latency whereby the estimation becomes a clear objective statistical procedure? Can this new statistical function contribute to the applicability and reliability of automated latency estimation? Is it possible to keep the computational time-cost low? In an attempt to answer these questions, we have developed a ‘double sliding-window’ technique with which to calculate the neuronal response onset latencies. The double sliding-window technique analyzes trial-by-trial data on PSTHs and combines the advantages of mathematical methods with the reliability of standard statistical processes. In order to check on the validity of the technique, we calculated the visual response onset latencies of neuronal responses obtained in a large number of extracellular single-unit recordings and compared them with visually quantified latencies and with the latencies provided by Poisson spike-train analysis, the CUSUM technique, and the advanced method of Falzett et al.
Section snippets
Results
For measurement of the latency of neuronal response onsets, we developed a software program, the double sliding-window technique, which slides two windows along the PSTHs. The first window (reference window) slides through the peristimulus period in 1 bin steps, and selects the portion that represents the maximum (or the minimum) frequency (depending on the excitatoric or inhibitoric characteristic of the neuronal response). A second window (sample window) then slides through, also in 1 bin
Discussion
The new method described here, the double sliding-window technique, allows the rapid, reproducible, accurate and automated estimation of the neuronal response onset latency. Our results show that the double sliding-window technique can yield more accurate latency data than Poisson spike-train analysis (Legéndy and Salcman, 1985), the CUSUM procedure (Ellaway, 1978) and the method of Falzett et al. (1985) in the sense that the EEs of the latencies calculated by the double sliding-window
Experimental procedures
The method presented here was specifically developed for the analysis of neuronal activity stored in PSTHs (Eördegh et al., 2005). The original temporal resolution of the recoded data was 1 ms, which was converted to a 5 ms binwidth for faster processing. Each PSTH consisted of a prestimulus interval and a peristimulus interval. The prestimulus interval contains the genuine spontaneous activity of a neuron, while the peristimulus interval contains both the spontaneous activity and the responses
Acknowledgments
The authors express their gratitude to Gabriella Dósai Molnár and Kálmán Hermann for their valuable technical assistance and to Péter Liszli for his expert help. The data-collecting activities of Zita Márkus and Zsuzsanna Paróczy are gratefully appreciated. This work was supported by OTKA/Hungary grants T 042610 and F 048396.
References (30)
- et al.
Splanchnic input to thoracic spinal neurons and its supraspinal modulation in the rat
Brain Res.
(1990) - et al.
A comparison of methods used to detect changes in neuronal discharge patterns
J. Neurosci. Methods
(1997) Cumulative sum technique and its application to the analysis of peristimulus time histograms
Electroencephalogr. Clin. Neurophysiol.
(1978)- et al.
Processing of spatial visual information along the pathway between the suprageniculate nucleus and the anterior ectosylvian cortex
Brain Res. Bull.
(2005) - et al.
A method for determining threshold from single-unit neural activity
Brain Res.
(1985) - et al.
An approach to the quantitative analysis of electrophysiological data from single neurons
Biophys. J.
(1960) - et al.
Temporal evolution of a decision-making process in medial premotor cortex
Neuron
(2002) - et al.
Effects of electrical and chemical stimulation of nucleus raphe magnus on responses to renal nerve stimulation
Brain Res.
(1991) - et al.
Spatial and temporal visual properties of single neurons in the suprageniculate nucleus of the thalamus
Neuroscience
(2006) - et al.
Task-related modulation in the monkey inferotemporal cortex
Brain Res.
(2006)
A statistical method for the estimation of neuronal response latency and its functional interpretation
Brain Res.
A cumulative sum test for a peri-stimulus time histogram using the Monte Carlo method
J. Neurosci. Methods
Speed-accuracy tradeoff in visual detection: applications of neural counting and timing
Vis. Res.
Retinal W-cell input to the upper superficial gray layer of the cat's superior colliculus: a conduction-velocity analysis
J. Neurophysiol.
The activity of monkey thalamic and motor cortical neurones in a skilled, ballistic movement
J. Physiol.
Cited by (9)
A review of the methods for neuronal response latency estimation
2015, BioSystemsCitation Excerpt :In statistical terms, this is a change-point estimation in the intensity (firing rate) function. A common technique involves binning the spike arrival times to form a peri-stimulus time histogram (PSTH) or more sophisticated estimates of the firing rate profile (Baker and Gerstein, 2001; Berényi et al., 2007; Chase and Young, 2008; Friedman and Priebe, 1998; Nawrot et al., 2003; Ventura, 2004). It is well known that many behavioral responses are completed so fast that the underlying sensory processes cannot rely on estimation of neural firing rate over an extended time window.
Visual stimulation synchronizes or desynchronizes the activity of neuron pairs between the caudate nucleus and the posterior thalamus
2011, Brain ResearchCitation Excerpt :Although latency studies are not widely used in the investigation of the thalamo-caudate relation, this investigation may provide additional information on the information flow between Sg and CN. In order to perform an objective statistical analysis of the onset response latencies, we used an automated method, the double sliding window method, developed in our laboratory (Berényi et al., 2007). We found no difference between the shortest latency values of the thalamic and the CN single-units.
Neuronal code of spatial visual information in the caudate nucleus
2011, NeuroscienceCitation Excerpt :Statistical analysis was performed with custom written scripts in the Matlab® software environment (MathWorks Inc., Natick, MA, USA). The visual response onset latencies were calculated using the double sliding windows method developed in our laboratory (for a detailed description see Berényi et al., 2007). The visual responses to moving visual noise stimulation of altogether 340 CN neurons were recorded.
Identification of noisy response latency
2012, Physical Review E - Statistical, Nonlinear, and Soft Matter PhysicsSynchrony makes neurons fire in sequence, and stimulus properties determine who is ahead
2011, Journal of NeuroscienceFeature extraction from spike trains with Bayesian binning: 'Latency is where the signal starts'
2010, Journal of Computational Neuroscience