A deep-learning approach for online cell identification and trace extraction in functional two-photon calcium imaging

Area of Science:

• Neuroscience
• Computational Neuroscience
• Bioimaging

Background:

• In vivo two-photon calcium imaging is vital for neuroscience but faces computational challenges in data processing.
• Online, frame-by-frame analysis of large field-of-view (FOV) imaging data remains difficult due to intensive computational demands.

Purpose of the Study:

• To introduce CITE-On (Cell Identification and Trace Extraction Online), a novel algorithm for rapid, automated analysis of two-photon calcium imaging data.
• To enable efficient online processing of neural activity from thousands of cells, including mesoscopic imaging.

Main Methods:

• Developed CITE-On, a convolutional neural network-based algorithm for cell identification, segmentation, identity tracking, and trace extraction.
• Utilized publicly available datasets for validation and performance comparison with existing state-of-the-art offline methods.

Main Results:

• CITE-On achieves performance comparable to state-of-the-art offline methods.
• The algorithm demonstrates generalization across various calcium indicators, brain regions, and imaging parameters in mice.
• Successfully processed thousands of cells online, extracting functional measurements from most neurons in the FOV.

Conclusions:

• CITE-On significantly accelerates image analysis for two-photon calcium imaging.
• This tool facilitates closed-loop neuroscience experiments, such as combined optical imaging and manipulation.