A matched-filter algorithm to detect amperometric spikes resulting from quantal secretion

Abstract

The MF approach performs well and leads into approaches to identify spike parameters.

Estimating amperometric spike parameters resulting from quantal exocytosis using curve fitting seeded by a matched-filter algorithm

“…The template-based matched filtering detection algorithm we recently described accurately detects spikes with few false positives (Balaji Ramachandran and Gillis, 2018). Here we show that the approach also generates initial parameter estimates useful to detect overlapping spikes and seed iterative curve fitting to extract complete information about the spike time course.…”

“…On the other hand, a more stringent criterion is applied to accept the second spike in a pair because the first spike must nearly completely decay to avoid disrupting the baseline for the second spike and allow foot analysis to proceed. Also note that the detection phase of analysis identifies essentially all spikes to allow accurate estimation of the frequency of exocytosis events near the electrode (Balaji Ramachandran and Gillis, 2018). However, only a subset of spikes can proceed to the analysis phase when events are detected at a high frequency because of overlap.…”

“…The templates used to detect spikes are a function consisting of a fast-exponential rise followed by a slower exponential decay: f(t)=(1−exp(−tτr)) exp (−tτd) Where, τ r is the rise time constant and τ d , is the decay time constants that are selected as representative of “typical” spike time courses. This least-squares fit of the template to the data (y(t)) is of the form, y^(t)=af(t) + b Where the amplitude a and baseline / offset b are found using the (non-iterative) least-squares algorithm (Balaji Ramachandran and Gillis, 2018). Our extended matched filtering approach fits spikes to a sum of templates with different decaying time constants in the form: y^(t)=a1f1(t) + a2f2(t) + b; t>0 Whereas least-square fits of templates to spikes enables excellent spike identification, the fits are imprecise because the time constants of the templates are fixed.…”

“…Use of the criterion score to determine a stable baseline region is, in some cases, preferable to measurement of standard deviation of the raw current because it is less sensitive to a sloping baseline (Fig. 4B in (Balaji Ramachandran and Gillis, 2018) and Supplemental Fig S1) or spurious interference, e.g., residual line noise.…”

“…The steps in amperometric spike analysis consist of spike detection followed by estimation of parameters. We previously described a matched filter (MF) template-based spike-detection algorithm that detects ~97% of manually identified spikes for a mere ~2% false positive rate (Balaji Ramachandran and Gillis, 2018). In brief, this algorithm uses a least-squares approach to fit prototypical spike templates consisting of a fast exponential rise and a slower exponential decay to each segment of data in a recording.…”