Temporal Deconvolution of Mesoscale Recordings

Abstract

1Mesoscale calcium recordings, such as wide-field imaging, enable high-temporal-resolution recordings of neuronal activity across extensive brain regions. However, because these recordings capture light emitted by calcium indicator fluorescence, the underlying neural activity is obscured by the temporal dynamics of the indicators. Here, we develop and evaluate four deconvolution methods to recover neuronal spiking rates from fluorescence traces recorded in wide-field imaging. The methods--Dynamically-Binning (using adaptive discrete-time bins), Continuously-Varying (estimating smooth spiking rates), First-Differences (providing efficient estimation), and a modified Wiener Filter (robust to large fluorescence magnitude variations)-consistently outperform the previously used adapted Lucy-Richardson algorithm on both synthetic data and simultaneous fluorescence-spike recordings. Critically, we demonstrate that using raw fluorescence signals in advanced analysis methods can yield distorted results, including spuriously inflated correlations between brain regions activity. This highlights the necessity of accurate temporal deconvolution for reliable interpretation of brain activity.