DLAO: a physics-informed deep learning framework for aberration correction in optical coherence tomography.

Area of Science:

• Biomedical Imaging
• Optical Engineering
• Deep Learning

Background:

• Optical coherence tomography (OCT) is crucial for high-resolution, non-invasive biomedical imaging.
• Image quality in OCT is often degraded by optical aberrations from system imperfections and sample inhomogeneities.
• These aberrations reduce spatial resolution and obscure fine details, limiting diagnostic and research applications.

Purpose of the Study:

• To develop an efficient framework for correcting complex aberrations in OCT images using deep learning and adaptive optics.
• To improve the spatial resolution and clarity of OCT images compromised by optical aberrations.

Main Methods:

• Introduced a physics-informed deep learning adaptive optics (DLAO) framework.
• Implemented a pseudo-point spread function (pseudo-PSF) preprocessing step to simplify aberration correction into a low-dimensional parameter estimation problem.
• Designed the layerwise adaptive progressive attention network (LAPANet) with multi-scale feature fusion and a LAPA module for hierarchical feature capture.

Main Results:

• The DLAO framework, particularly LAPANet, demonstrated superior performance in correcting aberrations compared to mainstream deep learning models.
• Achieved higher peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) scores.
• Maintained high inference efficiency and showed robustness and generalization capabilities in experiments.

Conclusions:

• The proposed DLAO framework effectively corrects complex aberrations in OCT images, significantly enhancing image quality.
• LAPANet's synergistic design precisely reconstructs critical image regions and restores high-frequency details.
• The framework offers practical value for improving OCT imaging in diagnostics and research.