Limits...
Novel use of matched filtering for synaptic event detection and extraction.

Shi Y, Nenadic Z, Xu X - PLoS ONE (2010)

Bottom Line: This new technique was applied to quantify and compare the EPSCs obtained from excitatory pyramidal cells and fast-spiking interneurons.In addition, our technique has been further applied to the detection and analysis of inhibitory postsynaptic current (IPSC) responses.Given the general purpose of our matched filtering and signal recognition algorithms, we expect that our technique can be appropriately modified and applied to detect and extract other types of electrophysiological and optical imaging signals.

View Article: PubMed Central - PubMed

Affiliation: Department of Anatomy and Neurobiology, School of Medicine, University of California Irvine, Irvine, California, USA.

ABSTRACT
Efficient and dependable methods for detection and measurement of synaptic events are important for studies of synaptic physiology and neuronal circuit connectivity. As the published methods with detection algorithms based upon amplitude thresholding and fixed or scaled template comparisons are of limited utility for detection of signals with variable amplitudes and superimposed events that have complex waveforms, previous techniques are not applicable for detection of evoked synaptic events in photostimulation and other similar experimental situations. Here we report on a novel technique that combines the design of a bank of approximate matched filters with the detection and estimation theory to automatically detect and extract photostimluation-evoked excitatory postsynaptic currents (EPSCs) from individually recorded neurons in cortical circuit mapping experiments. The sensitivity and specificity of the method were evaluated on both simulated and experimental data, with its performance comparable to that of visual event detection performed by human operators. This new technique was applied to quantify and compare the EPSCs obtained from excitatory pyramidal cells and fast-spiking interneurons. In addition, our technique has been further applied to the detection and analysis of inhibitory postsynaptic current (IPSC) responses. Given the general purpose of our matched filtering and signal recognition algorithms, we expect that our technique can be appropriately modified and applied to detect and extract other types of electrophysiological and optical imaging signals.

Show MeSH

Related in: MedlinePlus

Analysis of experimental data through matched filtering.A–F: Typical examples of detection and extraction of photostimulation-evoked EPSCs, in comparison with human visual detection. The raw data traces were shown in solid back, with the overlaying segments of EPSCs (blue) identified by an experienced human operator. The black crosses indicate the center of the human selected EPSCs. The color-coded segments shown below the raw data traces are detected and individually extracted EPSCs through matched filtering, with the respective crosses indicating the detected EPSC centers. The arrows and arrow heads point to the extra events correctly detected by the software, but missed by the human. The weak EPSCs (with the amplitudes of <20 pA, about the spontaneous EPSC level) (green) identified by the human are missed by the automated detection, because of the pre-set amplitude cutoff (20 pA). G: the bar graph summarizing the percentage of correct detection and percentage of false alarm of the automated detection of EPSCs across different data sets (N = 3), using the same filter bank at multiple detection thresholds (μ−1σ, μ−1.25σ and μ−1.5σ). The values are presented as mean ± SE. Each data set contained more than 200 photostimulation-evoked EPSCs, and the detection results were inspected and verified by a human operator.
© Copyright Policy
Related In: Results  -  Collection


getmorefigures.php?uid=PMC2991367&req=5

pone-0015517-g005: Analysis of experimental data through matched filtering.A–F: Typical examples of detection and extraction of photostimulation-evoked EPSCs, in comparison with human visual detection. The raw data traces were shown in solid back, with the overlaying segments of EPSCs (blue) identified by an experienced human operator. The black crosses indicate the center of the human selected EPSCs. The color-coded segments shown below the raw data traces are detected and individually extracted EPSCs through matched filtering, with the respective crosses indicating the detected EPSC centers. The arrows and arrow heads point to the extra events correctly detected by the software, but missed by the human. The weak EPSCs (with the amplitudes of <20 pA, about the spontaneous EPSC level) (green) identified by the human are missed by the automated detection, because of the pre-set amplitude cutoff (20 pA). G: the bar graph summarizing the percentage of correct detection and percentage of false alarm of the automated detection of EPSCs across different data sets (N = 3), using the same filter bank at multiple detection thresholds (μ−1σ, μ−1.25σ and μ−1.5σ). The values are presented as mean ± SE. Each data set contained more than 200 photostimulation-evoked EPSCs, and the detection results were inspected and verified by a human operator.

Mentions: Our new technique was further validated on experimental data, while compared to that of manual (human) detection. Typical examples of software detection and extraction of photostimulation-evoked EPSCs, along with human visual detection of these events, are illustrated in Figure 5 A–F. These data traces include direct responses and synaptically mediated EPSCs, and contain complex overlapping events. In most occasions, EPSCs detected by the software and the human operator matched quite well, with software detection performing better than the human in identification of overlapping synaptic events (see the arrow heads in Figure 5). It should be noted that some of the weak EPSCs (with the amplitudes of about the spontaneous EPSC level) identified by the human, however, were missed by the automated detection, because of the pre-set cutoff threshold for evoked EPSC amplitudes. The inclusion of the amplitude cut-off setting is necessary for rejecting noise-related artifacts, due to an inherent trade-off between the sensitivity and specificity of our and any other statistical detection method [15]. Those missed weak EPSCs were proportionally insignificant, as they accounted for less than 4% of all the candidate events across individual datasets. In addition, considering that the software detects both the spontaneous baseline synaptic activity and photostimulation responses, and as the baseline spontaneous response is subtracted from the photostimulation response, the missed measurement of weak EPSCs at the spontaneous level does not have a major impact on our final measurement and analysis of EPSCs across many photostimulation sites (data not shown). Figure 5G summarizes quantitative evaluations of the automated detection of EPSCs, using the same filter bank at multiple detection thresholds. In general, the method performance was excellent and stable across different data sets. With the detection results inspected and verified by experienced human operators, the average probability of correct detection (Pcd) is 87.7%, with the average false alarm (Pfa) rate of 2.6% for the three detection thresholds chosen as , and . Specifically, the probability of correct detection is 76.7%±2.4% (mean ± SE), 91.45%±3.1%, and 94.9%±2.43% respectively; the corresponding probability of false alarm is 0.67%±0.37%, 2.72%±0.44%, and 4.43%±1.75%, respectively. In practical settings, our software implementation includes quick tests of selected data traces to determine appropriate detection thresholds.


Novel use of matched filtering for synaptic event detection and extraction.

Shi Y, Nenadic Z, Xu X - PLoS ONE (2010)

Analysis of experimental data through matched filtering.A–F: Typical examples of detection and extraction of photostimulation-evoked EPSCs, in comparison with human visual detection. The raw data traces were shown in solid back, with the overlaying segments of EPSCs (blue) identified by an experienced human operator. The black crosses indicate the center of the human selected EPSCs. The color-coded segments shown below the raw data traces are detected and individually extracted EPSCs through matched filtering, with the respective crosses indicating the detected EPSC centers. The arrows and arrow heads point to the extra events correctly detected by the software, but missed by the human. The weak EPSCs (with the amplitudes of <20 pA, about the spontaneous EPSC level) (green) identified by the human are missed by the automated detection, because of the pre-set amplitude cutoff (20 pA). G: the bar graph summarizing the percentage of correct detection and percentage of false alarm of the automated detection of EPSCs across different data sets (N = 3), using the same filter bank at multiple detection thresholds (μ−1σ, μ−1.25σ and μ−1.5σ). The values are presented as mean ± SE. Each data set contained more than 200 photostimulation-evoked EPSCs, and the detection results were inspected and verified by a human operator.
© Copyright Policy
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC2991367&req=5

pone-0015517-g005: Analysis of experimental data through matched filtering.A–F: Typical examples of detection and extraction of photostimulation-evoked EPSCs, in comparison with human visual detection. The raw data traces were shown in solid back, with the overlaying segments of EPSCs (blue) identified by an experienced human operator. The black crosses indicate the center of the human selected EPSCs. The color-coded segments shown below the raw data traces are detected and individually extracted EPSCs through matched filtering, with the respective crosses indicating the detected EPSC centers. The arrows and arrow heads point to the extra events correctly detected by the software, but missed by the human. The weak EPSCs (with the amplitudes of <20 pA, about the spontaneous EPSC level) (green) identified by the human are missed by the automated detection, because of the pre-set amplitude cutoff (20 pA). G: the bar graph summarizing the percentage of correct detection and percentage of false alarm of the automated detection of EPSCs across different data sets (N = 3), using the same filter bank at multiple detection thresholds (μ−1σ, μ−1.25σ and μ−1.5σ). The values are presented as mean ± SE. Each data set contained more than 200 photostimulation-evoked EPSCs, and the detection results were inspected and verified by a human operator.
Mentions: Our new technique was further validated on experimental data, while compared to that of manual (human) detection. Typical examples of software detection and extraction of photostimulation-evoked EPSCs, along with human visual detection of these events, are illustrated in Figure 5 A–F. These data traces include direct responses and synaptically mediated EPSCs, and contain complex overlapping events. In most occasions, EPSCs detected by the software and the human operator matched quite well, with software detection performing better than the human in identification of overlapping synaptic events (see the arrow heads in Figure 5). It should be noted that some of the weak EPSCs (with the amplitudes of about the spontaneous EPSC level) identified by the human, however, were missed by the automated detection, because of the pre-set cutoff threshold for evoked EPSC amplitudes. The inclusion of the amplitude cut-off setting is necessary for rejecting noise-related artifacts, due to an inherent trade-off between the sensitivity and specificity of our and any other statistical detection method [15]. Those missed weak EPSCs were proportionally insignificant, as they accounted for less than 4% of all the candidate events across individual datasets. In addition, considering that the software detects both the spontaneous baseline synaptic activity and photostimulation responses, and as the baseline spontaneous response is subtracted from the photostimulation response, the missed measurement of weak EPSCs at the spontaneous level does not have a major impact on our final measurement and analysis of EPSCs across many photostimulation sites (data not shown). Figure 5G summarizes quantitative evaluations of the automated detection of EPSCs, using the same filter bank at multiple detection thresholds. In general, the method performance was excellent and stable across different data sets. With the detection results inspected and verified by experienced human operators, the average probability of correct detection (Pcd) is 87.7%, with the average false alarm (Pfa) rate of 2.6% for the three detection thresholds chosen as , and . Specifically, the probability of correct detection is 76.7%±2.4% (mean ± SE), 91.45%±3.1%, and 94.9%±2.43% respectively; the corresponding probability of false alarm is 0.67%±0.37%, 2.72%±0.44%, and 4.43%±1.75%, respectively. In practical settings, our software implementation includes quick tests of selected data traces to determine appropriate detection thresholds.

Bottom Line: This new technique was applied to quantify and compare the EPSCs obtained from excitatory pyramidal cells and fast-spiking interneurons.In addition, our technique has been further applied to the detection and analysis of inhibitory postsynaptic current (IPSC) responses.Given the general purpose of our matched filtering and signal recognition algorithms, we expect that our technique can be appropriately modified and applied to detect and extract other types of electrophysiological and optical imaging signals.

View Article: PubMed Central - PubMed

Affiliation: Department of Anatomy and Neurobiology, School of Medicine, University of California Irvine, Irvine, California, USA.

ABSTRACT
Efficient and dependable methods for detection and measurement of synaptic events are important for studies of synaptic physiology and neuronal circuit connectivity. As the published methods with detection algorithms based upon amplitude thresholding and fixed or scaled template comparisons are of limited utility for detection of signals with variable amplitudes and superimposed events that have complex waveforms, previous techniques are not applicable for detection of evoked synaptic events in photostimulation and other similar experimental situations. Here we report on a novel technique that combines the design of a bank of approximate matched filters with the detection and estimation theory to automatically detect and extract photostimluation-evoked excitatory postsynaptic currents (EPSCs) from individually recorded neurons in cortical circuit mapping experiments. The sensitivity and specificity of the method were evaluated on both simulated and experimental data, with its performance comparable to that of visual event detection performed by human operators. This new technique was applied to quantify and compare the EPSCs obtained from excitatory pyramidal cells and fast-spiking interneurons. In addition, our technique has been further applied to the detection and analysis of inhibitory postsynaptic current (IPSC) responses. Given the general purpose of our matched filtering and signal recognition algorithms, we expect that our technique can be appropriately modified and applied to detect and extract other types of electrophysiological and optical imaging signals.

Show MeSH
Related in: MedlinePlus