Soltanian-Zadeh, S. et al. Proc. Natl Acad. Sci. USA 116, 8554–8563 (2019).

Calcium imaging with two-photon microscopy is widely used to monitor neuronal activity in the brain. As the generated datasets are large, automated analysis methods are typically employed. Soltanian-Zadeh et al. developed a deep-learning approach to improve the accuracy of neuronal segmentation, which is the first step in automated processing pipelines. The Spatiotemporal NeuroNet (STNeuroNet) consists of a 3D convolutional neural network that can capture the spatiotemporal dynamics of calcium-based activity. To validate their approach and compare its performance to that of alternative tools, the researchers made use of publicly available datasets that they manually corrected for misattributed neurons. The researchers found that their method outperformed popular tools while maintaining a reasonable processing speed. The tool is available at https://github.com/soltanianzadeh/STNeuroNet.