Effective and Efficient Neural Networks for Spike Inference from In Vivo Calcium Imaging

Calcium imaging technique provides irreplaceable advantages in monitoring large population of neuronal activities simultaneously. However, due to the generally low signal to noise ratio (SNR) of the calcium signal and variability in dye properties, it is still challenging to faithfully infer neuronal spikes from these calcium signals, especially from in vivo experiments. In this study, we tackled the problem of both spike-rate and spike-event predictions using a data-driven approach, based on a public pool of dataset with simultaneously recorded calcium and electrophysiological signals using different dyes and recorded from different brain regions. We proposed the ENS2 (effective and efficient neural networks for spike inference from calcium signals) system using raw calcium inputs and it consistently outperforms state-of-the-arts algorithms in both spike-rate and spike-event predictions with reduced computational load. We have also demonstrated that factors such as sampling rates, smoothing window sizes and parametric evaluation metrics could readily bias the interpretation of inference performance. We concluded that optimizing our system for spike-event prediction could produce a more versatile inference system for real neuroscience studies.