Digital refocusing based on deep learning in optical coherence tomography.

Area of Science:

• Biomedical Optics
• Medical Imaging
• Artificial Intelligence

Background:

• Optical Coherence Tomography (OCT) provides high-resolution cross-sectional imaging but is limited by a shallow depth of focus.
• Extending the depth of focus in OCT is crucial for comprehensive volumetric analysis and disease diagnosis.
• Current methods for improving OCT imaging depth often involve complex hardware or post-processing techniques.

Purpose of the Study:

• To develop and validate a deep learning-based digital refocusing technique for extending the depth of focus in OCT.
• To enhance the lateral resolution of OCT images across a wider imaging range.
• To demonstrate the efficacy of the proposed method on both phantom and biological samples.

Main Methods:

• A generative adversarial network (GAN) architecture was employed, incorporating receptive field blocks.
• Pixel-level registered pairs of low-resolution (LR) and high-resolution (HR) OCT images were created from experimental data.
• The GAN was trained to learn the complex mapping between LR and HR OCT image pairs for digital refocusing.

Main Results:

• Significant improvement in lateral resolution was observed across a large imaging depth in OCT images.
• The digital refocusing approach effectively extended the depth of focus, providing clearer visualization.
• Successful demonstration on phantom and biological samples validates the method's practical applicability.

Conclusions:

• Deep learning-based digital refocusing is a promising approach to overcome the depth of focus limitations in OCT.
• This technique offers a powerful tool for optimizing OCT imaging quality and expanding its diagnostic capabilities.
• The study highlights the broad prospects of artificial intelligence in advancing optical coherence tomography.