Super-resolution of Retinal Optical Coherence Tomography Images Using Statistical Modeling

Area of Science:

• Medical Imaging
• Ophthalmology
• Image Processing

Background:

• Optical coherence tomography (OCT) is crucial for ophthalmology but suffers from noise and low resolution.
• Speckle noise and downsampling degrade OCT image quality, limiting diagnostic applications.

Purpose of the Study:

• To address the super-resolution (SR) challenge in retinal OCT images.
• To improve the quality of OCT images for enhanced diagnostic capabilities.

Main Methods:

• Utilized a Weibull mixture model (WMM) to characterize OCT intensity distribution.
• Applied expectation-maximization for WMM parameter estimation.
• Employed Gaussian transform and Gaussian mixture models for noise reduction.
• Implemented a patch-based algorithm with multivariate GMM prior and MAP estimator for super-resolution.

Main Results:

• The proposed method significantly suppresses noise in OCT images.
• Successfully reconstructs high-resolution (HR) images with improved visual quality.
• Outperforms existing super-resolution algorithms in terms of MSR and equivalent number of looks.

Conclusions:

• The developed statistical approach effectively enhances OCT image quality.
• The method is simple, requiring no special preprocessing.
• Improved image quality supports advanced OCT-assisted diagnostics.