Computational study demonstrates adaptive motion artifact correction in in vivo fluorescence imaging, indicating improved accuracy for quantitative neuroscience research.
Key Points
To develop a deep learning framework capable of adaptively correcting rigid and non-rigid physiological motion artifacts during in vivo fluorescence microscopy.
Constructed DeepMoCo, a deep learning architecture integrating graph neural networks to establish spatiotemporal correlations between biological motion and imaging artifacts.
Applied the framework to rapidly track and correct complex tissue dynamics and respiration-induced distortions in fluorescence imaging.
Achieved a 95% improvement in correlation coefficients relative to existing motion correction methods.
Demonstrated a 77% enhancement in peak signal-to-noise ratio compared to conventional techniques.