Mixed multiscale BM4D for three-dimensional optical coherence tomography denoising.

Area of Science:

• Biomedical Imaging
• Image Processing
• Computational Science

Background:

• Block matching and 4D filtering (BM4D) is a widely used image denoising technique.
• Existing BM4D methods face computational challenges when extended to 3D applications like optical coherence tomography (OCT).
• Denoising detail subbands in multiscale processing can be computationally intensive.

Purpose of the Study:

• To propose a computationally efficient multiscale extension of BM4D for 3D OCT image denoising.
• To simplify multiscale processing by avoiding direct denoising of detail subbands.
• To evaluate the performance of the proposed method against existing techniques and assess its impact on downstream tasks.

Main Methods:

• Developed a multiscale construction method in 2D and extended it to 3D.
• Proposed mixed multiscale BM4D (mmBM4D) specifically for OCT image denoising.
• Tested mmBM4D on three public OCT datasets from various imaging devices.

Main Results:

• mmBM4D significantly outperforms the original BM4D, showing improvements of over 0.68 dB in peak-signal-to-noise-ratio (PSNR) on one dataset.
• Achieved substantial enhancements in mean to standard deviation ratio, contrast to noise ratio, and equivalent number of looks on other datasets.
• Demonstrated competitive performance compared to state-of-the-art OCT denoising methods.

Conclusions:

• The proposed mmBM4D method offers an effective and computationally feasible approach for 3D OCT image denoising.
• mmBM4D preserves image quality, which is crucial for accurate downstream applications like retinal layer segmentation.
• This multiscale extension enhances the applicability of BM4D in complex 3D imaging scenarios.