Optical Coherence Tomography Noise Reduction Using Anisotropic Local Bivariate Gaussian Mixture Prior in 3D Complex...

Area of Science:

• Medical Imaging
• Signal Processing
• Optical Coherence Tomography (OCT)

Background:

• 3D OCT imaging is crucial for visualizing subsurface structures but is susceptible to noise, which degrades image quality and diagnostic accuracy.
• Traditional despeckling methods often struggle to preserve fine details and structural information in OCT data.
• Accurate statistical modeling of OCT data in the wavelet domain is essential for effective noise reduction.

Purpose of the Study:

• To develop an advanced noise-free data recovery technique for 3D OCT images.
• To improve the performance of Minimum Mean Square Error (MMSE) estimators by proposing a novel prior distribution for wavelet coefficients.
• To enhance the visual quality and diagnostic utility of 3D OCT data, particularly in the context of pathologies like wet Age-related Macular Degeneration (AMD).

Main Methods:

• Employed a Minimum Mean Square Error (MMSE) estimator in the 3D complex wavelet domain for OCT data recovery.
• Proposed a novel prior distribution for noise-free 3D complex wavelet coefficients, modeled as a mixture of two bivariate Gaussian probability density functions (PDFs) with local parameters.
• Utilized an anisotropic windowing procedure for local parameter estimation, tailored to the specific structure of OCT images, and explored various noise distributions (Gaussian/two-sided Rayleigh) and models (homomorphic/nonhomomorphic).

Main Results:

• The proposed bivariate Gaussian mixture prior effectively captures the heavy-tailed property and inter/intrascale dependencies of wavelet coefficients.
• Anisotropic windowing improved visual quality of the reconstructed OCT images.
• The optimal MMSE estimator, utilizing the local bivariate mixture prior with a nonhomomorphic model under Gaussian noise, demonstrated a significant improvement in Contrast-to-Noise Ratio (CNR) of 7.8 ± 1.7 dB on a 3D OCT dataset with wet AMD pathology.

Conclusions:

• The proposed MMSE-based approach with a bivariate Gaussian mixture prior offers a robust and effective solution for despeckling 3D OCT data.
• The method significantly enhances image quality, as evidenced by the substantial improvement in CNR, facilitating better visualization of pathologies.
• This technique holds promise for improving the diagnostic accuracy and clinical utility of 3D OCT imaging in ophthalmology and other fields.