Accurate detection of low signal-to-noise ratio neuronal calcium transient waves using a matched filter

Background: Calcium imaging has become a fundamental modality for studying neuronal circuit dynamics both in vitro and in vivo. However, identifying calcium events (CEs) from spectral data remains laborious and difficult, especially since the signal-to-noise ratio (SNR) often falls below 2. Existing automated signal detection methods are generally applied at high SNRs, leaving a large need for an automated algorithm that can accurately extract CEs from fluorescence intensity data of SNR 2 and below. New method: In this work we develop a Matched filter for Multi-unit Calcium Event (MMiCE) detection to extract CEs from fluorescence intensity traces of simulated and experimentally recorded neuronal calcium imaging data. Results: MMiCE reached perfect performance on simulated data with SNR >= 2 and a true positive (TP) rate of 98.27% (+/- 1.38% with a 95% confidence interval), and a false positive(FP) rate of 6.59% (+/- 2.56%) on simulated data with SNR 0.2. On real data, verified by patch-clamp recording, MMiCE performed with a TP rate of 100.00% (+/- 0.00) and a FP rate of 2.04% (+/- 4.10). Comparison with existing method(s): This high level of performance exceeds existing methods at SNRs as low as 0.2, which are well below those used in previous studies (SNR similar or equal to 5-10). Conclusion: Overall, the MMiCE detector performed exceptionally well on both simulated data, and experimentally recorded neuronal calcium imaging data. The MMiCE detector is accurate, reliable, well suited for wide-spread use, and freely available at sites.uci.edu/aggies or from the corresponding author. (C) 2015 Elsevier B.V. All rights reserved.