Abstract
Neurons emit axons, which form synapses, the fundamental unit of the nervous system. Neuroscientists use genetic anterograde tracing methods to label the synaptic output of specific neuronal subpopulations, but the resulting data sets are too large for manual analysis, and current automated methods have significant limitations in cost and quality. In this paper, we describe a pipeline optimized to identify anterogradely labeled presynaptic boutons in brain tissue sections. Our histologic pipeline labels boutons with high sensitivity and low background. To automatically detect labeled boutons in slide-scanned tissue sections, we developed BoutonNet. This detector uses a two-step approach: an intensity-based method proposes possible boutons, which are checked by a neural network-based confirmation step. BoutonNet was compared to expert annotation on a separate validation data set and achieved a result within human inter-rater variance. This open-source technique will allow quantitative analysis of the fundamental unit of the brain on a whole-brain scale.