MAS-Net OCT: a deep-learning-based speckle-free multiple aperture synthetic optical coherence tomography

Area of Science:

• Biomedical Optics
• Medical Imaging Technology
• Machine Learning in Healthcare

Background:

• Spectral domain optical coherence tomography (SD-OCT) faces a trade-off between transverse resolution and depth of focus (DOF).
• Speckle noise in OCT degrades image quality and limits resolution enhancement.
• Multiple aperture synthetic (MAS) OCT extends DOF by synthesizing a larger aperture.

Purpose of the Study:

• To develop a deep learning-based multiple aperture synthetic OCT (MAS-Net OCT) system.
• To integrate a self-supervised learning speckle-free model into MAS OCT.
• To evaluate the performance of MAS-Net OCT in improving resolution and reducing noise.

Main Methods:

• A novel deep learning framework, MAS-Net, was developed and trained on data from a MAS OCT system.
• The system integrates a self-supervised learning model for speckle reduction.
• Experiments were conducted on microparticle samples and various biological tissues.

Main Results:

• MAS-Net OCT demonstrated significant improvement in transverse resolution over a large imaging depth.
• The proposed method effectively reduced speckle noise in OCT images.
• Enhanced imaging quality was observed in both microparticle samples and biological tissues.

Conclusions:

• MAS-Net OCT successfully overcomes the resolution-DOF compromise inherent in SD-OCT.
• The integrated speckle-free model significantly improves image clarity.
• This deep learning approach offers a promising solution for advanced OCT imaging in clinical applications.