Reconstruction of Optical Coherence Tomography Images Using Mixed Low Rank Approximation and Second Order Tensor...

Area of Science:

• Medical Imaging
• Image Processing
• Biomedical Engineering

Background:

• Optical coherence tomography (OCT) images suffer from high noise and data subsampling.
• Effective super-resolution and denoising are crucial for reconstructing high-quality OCT images.
• Retinal layer preservation and artifact suppression are key challenges in OCT image processing.

Purpose of the Study:

• To propose a novel mixed low-rank approximation and second-order tensor-based total variation (LRSOTTV) approach.
• To enhance the super-resolution and denoising of retinal OCT images.
• To effectively utilize nonlocal spatial correlations and local smoothness properties for improved image reconstruction.

Main Methods:

• Constructed a third-order tensor from nonlocal similar 3D blocks using k-nearest neighbor grouping.
• Applied nuclear norm regularization to exploit nonlocal self-similarity.
• Introduced first-order tensor-based total variation (FOTTV) and second-order tensor-based total variation (SOTTV) regularization.
• Utilized the alternating direction method of multipliers (ADMM) for optimization.

Main Results:

• The LRSOTTV model demonstrated superior performance in both denoising and super-resolution compared to state-of-the-art methods.
• Integrating SOTTV provided noticeably improved results over FOTTV.
• The proposed method yielded OCT images with enhanced numerical and visual quality.

Conclusions:

• The proposed LRSOTTV approach effectively addresses noise and resolution issues in OCT imaging.
• The method successfully preserves fine details of retinal layers and suppresses artifacts.
• This technique offers a significant advancement for OCT image analysis and clinical applications.