Automated classification of signal sources in mesoscale calcium imaging

Abstract

Functional imaging of neural cell populations is critical for mapping intra- and inter-regional network dynamics across the neocortex. Recently we showed that an unsupervised machine learning decomposition of densely sampled recordings of cortical calcium dynamics results in a collection of components comprised of neuronal signal sources distinct from optical, movement, and vascular artifacts. Here we build a supervised learning classifier that automatically separates neural activity and artifact components, using a set of extracted spatial and temporal metrics that characterize the respective components. We demonstrate that the performance of the machine classifier matches human identification of signal components in novel data sets. Further, we analyze control data recorded in glial cell reporter and non-fluorescent mouse lines that validates human and machine identification of functional component class. This combined workflow of data-driven video decomposition and machine classification of signal sources will aid robust and scalable mapping of complex cerebral dynamics.