Axial super-resolution optical coherence tomography via complex-valued network.

Area of Science:

• Biomedical Imaging
• Optical Engineering
• Artificial Intelligence

Background:

• Optical Coherence Tomography (OCT) provides high-resolution, non-invasive cross-sectional images of biological tissues.
• Axial resolution in OCT, crucial for depth imaging, is limited by the light source bandwidth, impacting cost and performance.
• Real-valued deep learning has shown promise for optical super-resolution imaging.

Purpose of the Study:

• To develop a novel deep learning approach for axial super-resolution in OCT.
• To leverage both amplitude and phase information from OCT signals for improved imaging.
• To evaluate the performance and generalization capability of the proposed method.

Main Methods:

• Proposed a complex-valued super-resolution network (CVSR-Net) specifically designed for OCT data.
• Utilized the full complex-valued OCT signal, including amplitude and phase, within the neural network architecture.
• Evaluated CVSR-Net on three distinct OCT datasets and compared it against real-valued and complex-valued network counterparts.

Main Results:

• CVSR-Net demonstrated superior axial super-resolution performance compared to its real-valued counterpart.
• The proposed complex-valued network outperformed six prevailing real-valued networks and their complex-valued versions.
• CVSR-Net maintained strong super-resolution performance on out-of-distribution datasets, indicating robust generalization.

Conclusions:

• Complex-valued deep learning effectively enhances axial resolution in OCT imaging.
• CVSR-Net offers a significant advancement in OCT super-resolution by utilizing complete signal information.
• The proposed method shows potential for broader applications in biomedical imaging due to its generalization capability.