Temporal Deconvolution of Mesoscale Recordings

Mesoscale calcium imaging techniques, such as wide-field imaging, enable high temporal resolution recordings of extensive neuronal activity across one or more brain regions. However, since the recordings capture light emission generated by the fluorescence of the calcium indicator, the neural activity that drives the calcium changes is masked by the dynamics of the calcium indicator. In this study, we develop and evaluate new methods to deconvolve fluorescence traces into the underlying neuronal spiking rates driving them. Our new inference methods take into account both the noise in the recordings and the temporal dynamics of the calcium indicator response. Our first proposed method, termed Dynamical-Binning, estimates spiking rates that are constant over discrete time bins. The size of each time bin depends on the data and is determined dynamically. Our second method, Continuously-Varying, estimates the spiking rate as a continuous function. This method aims at studies seeking to find slow rate fluctuations rather than identifying abrupt changes in the spiking rate. The third method, First-Differences, aims to give a quick estimate of the spiking rates, which is beneficial for exceptionally large datasets, typical of mesoscale recordings. Our fourth method, is a modified Weiner Filter. It estimates spiking rates by efficiently removing noise with a fixed ratio compared to the signal. This approach is beneficial for datasets exhibiting large fluctuations in fluorescence magnitudes. We compare the accuracy of our methods against the existing Lucy-Richardson image recovery algorithm in its adapted form to recover temporal dynamics. Our results demonstrate that all our proposed methods surpass the performance of Lucy-Richardson on both synthetic and recording datasets, including concurrent recordings of fluorescence and spike counts from the exact origin by multichannel silicon probes. Furthermore, we illustrate that our findings are indifferent to the choice of removing hemodynamic signals. Lastly, we demonstrate that the reliance on calcium signals for advanced analytical approaches can lead to distorted results. For example, they can show correlations between the activity of different brain regions that are unrealistically high compared to the correlations of the underlying spiking rate between the same areas. This highlights the critical importance of temporal inference for further accurate and reliable analysis in understanding the complexities of brain activity.