Image restoration for blurry optical images caused by photon diffusion with deep learning

Area of Science:

• Biomedical Optics
• Medical Imaging
• Deep Learning Applications

Background:

• Optical macroscopic imaging is crucial for biomedical investigations, providing structural and functional insights.
• Photon diffusion in biological tissues significantly degrades spatial resolution in optical imaging.
• Limited resolution restricts the application of optical imaging in deep tissue visualization.

Purpose of the Study:

• To develop an image restoration method for optical macroscopic imaging using deep learning.
• To address the challenge of poor spatial resolution caused by photon diffusion.
• To improve the detection of small targets at deep locations within biological tissues.

Main Methods:

• Proposed a deep learning-based image restoration method for optical macroscopic imaging.
• Utilized two blurry images captured at orthogonal angles to ensure solution uniqueness.
• Developed a fully convolutional neural network with V-shaped networks and a synthetical path.

Main Results:

• The proposed method effectively eliminates blur caused by photon diffusion.
• Restored images showed improved clarity and detail, particularly for small, deep targets.
• Simulations, phantom, and mouse experiments validated the method's effectiveness.

Conclusions:

• Deep learning offers a powerful solution for enhancing optical macroscopic imaging resolution.
• The dual-angle imaging approach combined with a specialized neural network improves deep tissue visualization.
• This technique has significant potential for advancing biomedical imaging applications.