Limits...
Waveform Similarity Analysis: A Simple Template Comparing Approach for Detecting and Quantifying Noisy Evoked Compound Action Potentials.

Potas JR, de Castro NG, Maddess T, de Souza MN - PLoS ONE (2015)

Bottom Line: Signals were detected and quantified using Waveform Similarity Analysis, which was compared to event detection, latency and magnitude measurements of the same signals performed by a trained observer, a process we called Trained Eye Analysis.Compared to the trained eye, Waveform Similarity Analysis is automatic, objective, does not rely on the observer to identify and/or measure peaks, and can detect small clustered events even when signal-to-noise ratio is poor.Waveform Similarity Analysis provides a simple, reliable and convenient approach to quantify latencies and magnitudes of complex waveforms and therefore serves as a useful tool for studying evoked compound action potentials in neural regeneration studies.

View Article: PubMed Central - PubMed

Affiliation: Department of Neuroscience, John Curtin School of Medical Research, Australian National University, Canberra, ACT, Australia; Medical School, Australian National University, Canberra, ACT, Australia; Instituto de Ciências Biomédicas, Universidade Federal do Rio de Janeiro, Rio de Janeiro, RJ, Brazil.

ABSTRACT
Experimental electrophysiological assessment of evoked responses from regenerating nerves is challenging due to the typical complex response of events dispersed over various latencies and poor signal-to-noise ratio. Our objective was to automate the detection of compound action potential events and derive their latencies and magnitudes using a simple cross-correlation template comparison approach. For this, we developed an algorithm called Waveform Similarity Analysis. To test the algorithm, challenging signals were generated in vivo by stimulating sural and sciatic nerves, whilst recording evoked potentials at the sciatic nerve and tibialis anterior muscle, respectively, in animals recovering from sciatic nerve transection. Our template for the algorithm was generated based on responses evoked from the intact side. We also simulated noisy signals and examined the output of the Waveform Similarity Analysis algorithm with imperfect templates. Signals were detected and quantified using Waveform Similarity Analysis, which was compared to event detection, latency and magnitude measurements of the same signals performed by a trained observer, a process we called Trained Eye Analysis. The Waveform Similarity Analysis algorithm could successfully detect and quantify simple or complex responses from nerve and muscle compound action potentials of intact or regenerated nerves. Incorrectly specifying the template outperformed Trained Eye Analysis for predicting signal amplitude, but produced consistent latency errors for the simulated signals examined. Compared to the trained eye, Waveform Similarity Analysis is automatic, objective, does not rely on the observer to identify and/or measure peaks, and can detect small clustered events even when signal-to-noise ratio is poor. Waveform Similarity Analysis provides a simple, reliable and convenient approach to quantify latencies and magnitudes of complex waveforms and therefore serves as a useful tool for studying evoked compound action potentials in neural regeneration studies.

No MeSH data available.


Related in: MedlinePlus

Simulation experiment of WSA performance under conditions of template distortion.Simulated signals generated by adding three multiples (0.5×, 1.0× and 2.0×) of a signal of interest (A, true signal) to noise (B), were used to derive a signal for analysis (C, total). WSA was applied using three templates; the correct one identical to 1.0× the true signal (C, blue), a template with double (C, red) and half (C, green) the frequency content of the correct template. Latency-corrected WSA outputs calculated from the cross-correlation sequence (xycorr) are shown (D) for each template (WSA1, blue, output with correct template; WSA2, red, template with double the correct frequency; WSA3, green, template with half the correct frequency). Overlayed to WSA outputs is the true signal, with the positive and negative peaks indicated (black dots) that reports the true peak-to-peak measurements for comparing the performance of WSA and TEA. Peak detection threshold at >4× the standard deviation of the first 400 ms of each respective WSA output was used to automate peak detection (D, blue, red and green dots indicate peaks for WSA1, WSA2 and WSA3 respectively). The latency (E) and Magnitude (F) errors are shown for WSA performed by each template (following latency and amplitude correction respectively) and TEA for the 6 samples of each signal. The black line indicates zero error; data expressed as mean ± SEM.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4556619&req=5

pone.0136992.g005: Simulation experiment of WSA performance under conditions of template distortion.Simulated signals generated by adding three multiples (0.5×, 1.0× and 2.0×) of a signal of interest (A, true signal) to noise (B), were used to derive a signal for analysis (C, total). WSA was applied using three templates; the correct one identical to 1.0× the true signal (C, blue), a template with double (C, red) and half (C, green) the frequency content of the correct template. Latency-corrected WSA outputs calculated from the cross-correlation sequence (xycorr) are shown (D) for each template (WSA1, blue, output with correct template; WSA2, red, template with double the correct frequency; WSA3, green, template with half the correct frequency). Overlayed to WSA outputs is the true signal, with the positive and negative peaks indicated (black dots) that reports the true peak-to-peak measurements for comparing the performance of WSA and TEA. Peak detection threshold at >4× the standard deviation of the first 400 ms of each respective WSA output was used to automate peak detection (D, blue, red and green dots indicate peaks for WSA1, WSA2 and WSA3 respectively). The latency (E) and Magnitude (F) errors are shown for WSA performed by each template (following latency and amplitude correction respectively) and TEA for the 6 samples of each signal. The black line indicates zero error; data expressed as mean ± SEM.

Mentions: To examine the effect of applying WSA with an imperfect template, a simulation experiment was conducted (Fig 5) which used a simple sine wave “true” signal (Fig 5A) with zero-mean normally distributed random noise (Fig 5B) added to generate a “total” signal for analysis (Fig 5C) that represents a typical signal acquired under experimental conditions. Three amplitudes of the true signal were used (0.5×, 1.0× and 2.0×) to examine the performance of WSA under various template conditions. These conditions included the correct template (identical to 1.0× the true signal), and 16 distorted templates ranging from half to double the frequency content of the correct template, which were created by expanding or contracting the original template in time. The latency and magnitudes were automatically detected using 4× the standard deviation of the respective WSA output on a background region of the cross-correlation sequence (i.e. the first 400 ms prior to the first signal of interest). The WSA outputs for each template were the event latency (corrected) and the magnitude (amplitude-corrected to 1.0× true signal: mean peak-to-peak of true signal/mean WSA output). These were then compared to TEA. TEA was performed on a separate figure that only presented the total signal, so that the observer was blind to the location of the true signal. Quantification was performed on 6 samples for each signal magnification (Fig 5 shows examples of two for each signal) and mean and SEM calculated for comparisons. Errors in latency (latency of true signal minus latency determined by WSA or TEA) and magnitude (magnitude of true signal minus magnitude generated by WSA or TEA, expressed as a percent of the magnitude of the true signal) were calculated and expressed as mean ± SEM. See Fig 5A–5D for further details.


Waveform Similarity Analysis: A Simple Template Comparing Approach for Detecting and Quantifying Noisy Evoked Compound Action Potentials.

Potas JR, de Castro NG, Maddess T, de Souza MN - PLoS ONE (2015)

Simulation experiment of WSA performance under conditions of template distortion.Simulated signals generated by adding three multiples (0.5×, 1.0× and 2.0×) of a signal of interest (A, true signal) to noise (B), were used to derive a signal for analysis (C, total). WSA was applied using three templates; the correct one identical to 1.0× the true signal (C, blue), a template with double (C, red) and half (C, green) the frequency content of the correct template. Latency-corrected WSA outputs calculated from the cross-correlation sequence (xycorr) are shown (D) for each template (WSA1, blue, output with correct template; WSA2, red, template with double the correct frequency; WSA3, green, template with half the correct frequency). Overlayed to WSA outputs is the true signal, with the positive and negative peaks indicated (black dots) that reports the true peak-to-peak measurements for comparing the performance of WSA and TEA. Peak detection threshold at >4× the standard deviation of the first 400 ms of each respective WSA output was used to automate peak detection (D, blue, red and green dots indicate peaks for WSA1, WSA2 and WSA3 respectively). The latency (E) and Magnitude (F) errors are shown for WSA performed by each template (following latency and amplitude correction respectively) and TEA for the 6 samples of each signal. The black line indicates zero error; data expressed as mean ± SEM.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4556619&req=5

pone.0136992.g005: Simulation experiment of WSA performance under conditions of template distortion.Simulated signals generated by adding three multiples (0.5×, 1.0× and 2.0×) of a signal of interest (A, true signal) to noise (B), were used to derive a signal for analysis (C, total). WSA was applied using three templates; the correct one identical to 1.0× the true signal (C, blue), a template with double (C, red) and half (C, green) the frequency content of the correct template. Latency-corrected WSA outputs calculated from the cross-correlation sequence (xycorr) are shown (D) for each template (WSA1, blue, output with correct template; WSA2, red, template with double the correct frequency; WSA3, green, template with half the correct frequency). Overlayed to WSA outputs is the true signal, with the positive and negative peaks indicated (black dots) that reports the true peak-to-peak measurements for comparing the performance of WSA and TEA. Peak detection threshold at >4× the standard deviation of the first 400 ms of each respective WSA output was used to automate peak detection (D, blue, red and green dots indicate peaks for WSA1, WSA2 and WSA3 respectively). The latency (E) and Magnitude (F) errors are shown for WSA performed by each template (following latency and amplitude correction respectively) and TEA for the 6 samples of each signal. The black line indicates zero error; data expressed as mean ± SEM.
Mentions: To examine the effect of applying WSA with an imperfect template, a simulation experiment was conducted (Fig 5) which used a simple sine wave “true” signal (Fig 5A) with zero-mean normally distributed random noise (Fig 5B) added to generate a “total” signal for analysis (Fig 5C) that represents a typical signal acquired under experimental conditions. Three amplitudes of the true signal were used (0.5×, 1.0× and 2.0×) to examine the performance of WSA under various template conditions. These conditions included the correct template (identical to 1.0× the true signal), and 16 distorted templates ranging from half to double the frequency content of the correct template, which were created by expanding or contracting the original template in time. The latency and magnitudes were automatically detected using 4× the standard deviation of the respective WSA output on a background region of the cross-correlation sequence (i.e. the first 400 ms prior to the first signal of interest). The WSA outputs for each template were the event latency (corrected) and the magnitude (amplitude-corrected to 1.0× true signal: mean peak-to-peak of true signal/mean WSA output). These were then compared to TEA. TEA was performed on a separate figure that only presented the total signal, so that the observer was blind to the location of the true signal. Quantification was performed on 6 samples for each signal magnification (Fig 5 shows examples of two for each signal) and mean and SEM calculated for comparisons. Errors in latency (latency of true signal minus latency determined by WSA or TEA) and magnitude (magnitude of true signal minus magnitude generated by WSA or TEA, expressed as a percent of the magnitude of the true signal) were calculated and expressed as mean ± SEM. See Fig 5A–5D for further details.

Bottom Line: Signals were detected and quantified using Waveform Similarity Analysis, which was compared to event detection, latency and magnitude measurements of the same signals performed by a trained observer, a process we called Trained Eye Analysis.Compared to the trained eye, Waveform Similarity Analysis is automatic, objective, does not rely on the observer to identify and/or measure peaks, and can detect small clustered events even when signal-to-noise ratio is poor.Waveform Similarity Analysis provides a simple, reliable and convenient approach to quantify latencies and magnitudes of complex waveforms and therefore serves as a useful tool for studying evoked compound action potentials in neural regeneration studies.

View Article: PubMed Central - PubMed

Affiliation: Department of Neuroscience, John Curtin School of Medical Research, Australian National University, Canberra, ACT, Australia; Medical School, Australian National University, Canberra, ACT, Australia; Instituto de Ciências Biomédicas, Universidade Federal do Rio de Janeiro, Rio de Janeiro, RJ, Brazil.

ABSTRACT
Experimental electrophysiological assessment of evoked responses from regenerating nerves is challenging due to the typical complex response of events dispersed over various latencies and poor signal-to-noise ratio. Our objective was to automate the detection of compound action potential events and derive their latencies and magnitudes using a simple cross-correlation template comparison approach. For this, we developed an algorithm called Waveform Similarity Analysis. To test the algorithm, challenging signals were generated in vivo by stimulating sural and sciatic nerves, whilst recording evoked potentials at the sciatic nerve and tibialis anterior muscle, respectively, in animals recovering from sciatic nerve transection. Our template for the algorithm was generated based on responses evoked from the intact side. We also simulated noisy signals and examined the output of the Waveform Similarity Analysis algorithm with imperfect templates. Signals were detected and quantified using Waveform Similarity Analysis, which was compared to event detection, latency and magnitude measurements of the same signals performed by a trained observer, a process we called Trained Eye Analysis. The Waveform Similarity Analysis algorithm could successfully detect and quantify simple or complex responses from nerve and muscle compound action potentials of intact or regenerated nerves. Incorrectly specifying the template outperformed Trained Eye Analysis for predicting signal amplitude, but produced consistent latency errors for the simulated signals examined. Compared to the trained eye, Waveform Similarity Analysis is automatic, objective, does not rely on the observer to identify and/or measure peaks, and can detect small clustered events even when signal-to-noise ratio is poor. Waveform Similarity Analysis provides a simple, reliable and convenient approach to quantify latencies and magnitudes of complex waveforms and therefore serves as a useful tool for studying evoked compound action potentials in neural regeneration studies.

No MeSH data available.


Related in: MedlinePlus