Fast and robust active neuron segmentation in two-photon calcium imaging using spatiotemporal deep learning.

Area of Science:

• Neuroscience
• Computational Biology
• Biophysics

Background:

• Calcium imaging enables large-scale in vivo neuronal activity recording at cellular resolution.
• Accurate and efficient active neuron segmentation is crucial for real-time analysis and discovery of neuronal coding properties.

Purpose of the Study:

• To develop a 3D convolutional neural network for automated, fast, and reliable identification and segmentation of active neurons in two-photon calcium imaging movies.
• To address limitations in current neuron segmentation techniques and improve the analysis of neuronal signals.

Main Methods:

• A 3D convolutional neural network architecture was employed to process spatiotemporal information from two-photon calcium imaging data.
• The method was validated using diverse two-photon microscopy datasets, including data from different cortical layers with varying neuron densities.

Main Results:

• The proposed 3D convolutional neural network method demonstrated superior performance compared to state-of-the-art techniques.
• Segmentation accuracy was comparable to manual segmentation, highlighting its reliability.
• The network showed robustness, performing accurately across different cortical layers.
• Significant tabulation flaws were identified in a prominent online neuron segmentation challenge.

Conclusions:

• The developed 3D convolutional neural network offers a computationally fast and accurate solution for active neuron segmentation in calcium imaging.
• This tool is valuable for real-time optogenetic experiments and advancing the understanding of neuronal activity.
• Open-source software and an annotated dataset are provided to facilitate further research.