Abstract
Information transfer and integration in the brain occurs at chemical synapses and is mediated by the fusion of synaptic vesicles filled with neurotransmitter. Synaptic vesicle dynamic spatial organization regulates synaptic transmission as well as synaptic plasticity. Because of their small size, synaptic vesicles require electron microscopy (EM) for their imaging, and their analysis is conducted manually. The manual annotation and segmentation of the hundreds to thousands of synaptic vesicles, is highly time consuming and limits the throughput of data collection. To overcome this limitation, we built an algorithm, mainly relying on convolutional neural networks (CNNs), capable of automatically detecting and localizing synaptic vesicles in electron micrographs. The algorithm was trained on murine synapses but we show that it works well on synapses from different species, ranging from zebrafish to human, and from different preparations. As output, we provide the vesicle count and coordinates, the nearest neighbor distance (nnd) and the estimate of the vesicles area. We also provide a graphical user interface (GUI) to guide users through image analysis, result visualization, and manual proof-reading. The application of our algorithm is especially recommended for images produced by transmission EM. Since this type of imaging is used routinely to investigate presynaptic terminals, our solution will likely be of interest for numerous research groups.