Tri-view two-photon microscopic image registration and deblurring with convolutional neural networks.

Area of Science:

• Neuroscience
• Biomedical Imaging
• Computational Biology

Background:

• Two-photon fluorescence microscopy allows 3D neural imaging of deep cortical regions.
• Image quality in the depth (z-axis) is often degraded by lens blur, hindering neural connectivity analysis.

Purpose of the Study:

• To develop a novel method for restoring isotropic image volumes in 3D neural microscopy.
• To improve the identification of neural connectivity by enhancing image quality along the depth direction.

Main Methods:

• A novel approach using cascaded convolutional neural networks (CNNs) for image restoration.
• The method employs rigid transformation, dense registration, and deblurring networks for efficient processing.
• CNN models were trained using self-supervised learning with simulated microscopic images accounting for imaging distortions.

Main Results:

• Substantial improvements in 3D neural image quality were achieved.
• The proposed method effectively restores isotropic image volumes by fusing orthogonal viewpoints.
• Enhanced image clarity facilitates better visualization of neural structures and connectivity.

Conclusions:

• The developed CNN-based approach significantly enhances 3D neural imaging resolution and clarity.
• This method offers a powerful tool for detailed neural connectivity studies in neuroscience.
• The self-supervised learning strategy with simulated data enables robust model training for microscopy image restoration.