DeepVID v2: Self-Supervised Denoising with Decoupled Spatiotemporal Enhancement for Low-Photon Voltage Imaging

Area of Science:

• Neuroscience
• Biophysics
• Computer Science

Background:

• Voltage imaging is crucial for studying neuronal activity but suffers from severe Poisson noise in low-photon conditions, hindering accurate analysis.
• Existing self-supervised deep learning methods for denoising voltage imaging data often face a trade-off between spatial and temporal performance.

Approach:

• DeepVID v2 is a novel self-supervised denoising framework designed to enhance low-photon voltage imaging.
• It builds upon the original DeepVID framework, incorporating frame-based denoising with temporal information leverage and multiple blind pixel integration for spatial details.
• A new edge extraction branch is introduced to capture fine structural details and learn high spatial resolution information.

Key Points:

• DeepVID v2 effectively overcomes the spatial-temporal performance trade-off in denoising voltage imaging data.
• The framework achieves superior denoising, enabling the resolution of high-resolution spatial structures and rapid neuronal activities.
• DeepVID v2 demonstrates generalization across various imaging conditions, including different signal-to-noise ratios and extreme low-photon scenarios.

Conclusions:

• DeepVID v2 is a powerful tool for enhancing voltage imaging, offering significant improvements in data quality.
• The framework shows potential for broader application to other low-photon imaging modalities.
• DeepVID v2 can greatly facilitate the study of neuronal activities in the brain.