Data-driven segmentation of cortical calcium dynamics

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biophysics

Background:

• Mapping cortical organization requires accurate demixing of neuronal calcium signals from transcranial videos.
• Current methods face challenges in separating neural activity from artifacts in complex cerebral dynamics.

Purpose of the Study:

• To develop and validate a method for optimal recovery of neural signal content from widefield calcium imaging data.
• To establish a data-driven approach for functional segmentation of the mouse cortex.

Main Methods:

• Utilized independent component analysis (ICA) for signal demixing in widefield neuronal calcium recordings.
• Developed a random forest classifier using spatial and temporal metrics to distinguish neural activity from artifacts.
• Applied the workflow to generate functional segmentation maps of the mouse cerebral cortex.

Main Results:

• ICA optimally recovered neural signal content at a minimum sampling rate of 1.5×10^6 pixels/frame.
• The random forest classifier achieved human-level performance in separating neural and artifact components.
• Generated functional segmentation maps identifying approximately 115 domains per hemisphere.
• Revealed distinct regional motifs in domain morphology, with higher-order regions showing larger, eccentric domains.

Conclusions:

• The demonstrated workflow of data-driven video decomposition and machine classification enhances high-quality mapping of cerebral dynamics.
• This approach provides a robust method for functional segmentation of the mouse cortex.
• The findings contribute to a deeper understanding of cortical organization and neural signal processing.