maskNMF: A denoise-sparsen-detect approach for extracting neural signals from dense imaging data

Area of Science:

• Neuroscience
• Computational Biology
• Imaging Science

Background:

• Calcium imaging is crucial for monitoring large neuronal populations.
• Bessel imaging offers high-density neural recordings but requires effective signal demixing.
• Existing demixing methods struggle with high imaging densities.

Approach:

• Introduced maskNMF, a novel computational pipeline for demixing dense calcium imaging data.
• Denoising and temporal sparsening enhance signal strength and reduce spatial overlap.
• A neural network, trained on electron microscopy-derived shapes, detects neurons without manual selection.
• Constrained non-negative matrix factorization demixes activity, initialized with detected cell shapes.

Key Points:

• maskNMF achieves accurate demixing on denser datasets than previously possible.
• The pipeline successfully processes both simulated and real calcium imaging data.
• It also performs well on standard two-photon imaging data.
• The algorithm is highly parallelizable and processes data faster than real time.

Conclusions:

• maskNMF enables faithful imaging of larger neural populations by overcoming limitations in demixing dense data.
• This computational approach significantly advances the capacity to study neural circuits.
• The speed and accuracy of maskNMF make it a valuable tool for neuroscience research.