Blind sparse deconvolution for inferring spike trains from fluorescence recordings

Abstract

The parallel developments of genetically-encoded calcium indicators and fast fluorescence imaging techniques makes it possible to simultaneously record neural activity of extended neuronal populations in vivo, opening a new arena for systems neuroscience. To fully harness the potential of functional imaging, one needs to infer the sequence of action potentials from fluorescence time traces. Here we build on recently proposed computational approaches to develop a blind sparse deconvolution algorithm (BSD), which we motivate by a theoretical analysis. We demonstrate that this method outperforms existing sparse deconvolution algorithms in terms of robustness, speed and/or accuracy on both synthetic and real fluorescence data. Furthermore, we provide solutions for the practical problems of thresholding and determination of the rise and decay time constants. We provide theoretical bounds on the performance of the algorithm in terms of precision-recall and temporal accuracy. Finally, we extend the computational framework to support temporal super-resolution whose performance is established on real data.