Temporal and volumetric denoising via quantile sparse image prior.

Area of Science:

• Medical Imaging
• Image Processing
• Computational Biology

Background:

• Optical coherence tomography (OCT) provides high-resolution retinal scans but suffers from speckle noise.
• Computed tomography (CT) offers lower resolution with high-frequency noise, posing distinct denoising challenges.
• Existing denoising methods may not universally preserve image structures across diverse medical imaging modalities.

Purpose of the Study:

• To introduce a universal and structure-preserving regularization term, the quantile sparse image (QuaSI) prior.
• To develop a variational framework utilizing the QuaSI prior and a Huber data fidelity model for 3-D and 3-D+t image denoising.
• To demonstrate the effectiveness of the proposed method on volumetric OCT and CT data.

Main Methods:

• Development of the quantile sparse image (QuaSI) prior as a regularization term.
• Implementation of a variational framework with a Huber data fidelity model.
• Application of an alternating direction method of multipliers (ADMM) for efficient optimization, including linearization of the quantile filter.

Main Results:

• The QuaSI prior proved effective for denoising both OCT and CT data, which exhibit different noise characteristics.
• The proposed variational framework successfully handled 3-D and 3-D+t data.
• Experimental results on multiple datasets confirmed the excellent performance of the QuaSI-based denoising method.

Conclusions:

• The proposed QuaSI prior offers a universal and structure-preserving approach to medical image denoising.
• The developed variational framework with ADMM optimization is efficient and effective for various medical imaging data.
• This method shows significant potential for improving image quality in OCT, CT, and other modalities.