Tensor Ring Decomposition Guided Dictionary Learning for OCT Image Denoising.

Area of Science:

• Biomedical Imaging
• Medical Image Processing
• Ophthalmology

Background:

• Optical coherence tomography (OCT) is vital for retinal imaging but suffers from speckle noise.
• Speckle noise obscures critical morphological details, hindering accurate diagnosis of ocular diseases.

Purpose of the Study:

• To introduce a novel Tensor Ring Decomposition-guided Dictionary Learning (TRGDL) model for OCT image denoising.
• To effectively utilize 3D low-rank and sparsity priors within a unified framework.

Main Methods:

• Constructing OCT group tensors from cubic patches and clustering similar patches.
• Applying Tensor Ring (TR) decomposition to exploit low-rank structure within OCT group tensors.
• Learning shared dictionaries in spatial and temporal dimensions to leverage inter-group sparsity.
• Developing an optimization algorithm using proximal alternating minimization and ADMM.

Main Results:

• The TRGDL model successfully denoises OCT images by integrating spatial, non-local, and temporal correlations.
• Experiments demonstrate superior performance of TRGDL over existing methods for OCT image denoising.
• Qualitative and quantitative evaluations confirm the model's effectiveness across diverse OCT datasets.

Conclusions:

• The proposed TRGDL model offers a powerful and generalizable solution for OCT image denoising.
• TRGDL significantly improves image quality, aiding in more precise clinical diagnosis of eye conditions.
• This approach advances the application of advanced signal processing techniques in medical imaging.