RFDNet: Robust Frequency-Based Denoising Network for 3D Ultrasound Vascular Imaging Using a Row-Column Addressed...

Area of Science:

• Medical Imaging
• Ultrasound Technology
• Artificial Intelligence in Medicine

Background:

• Three-dimensional ultrasound vascular imaging (3D UVI) is crucial for visualizing complex vascular structures.
• Row-column addressed (RCA) arrays, common in 3D UVI, introduce ramp-shaped noise due to point spread function (PSF) anisotropy, degrading image quality.
• Existing denoising methods struggle with domain shift bias, specific data needs, and inter-slice inconsistencies from 2D slice-wise training.

Purpose of the Study:

• To develop a novel denoising method, Robust Frequency-based Denoising Network (RFDNet), to overcome limitations in 3D UVI.
• To suppress ramp-shaped noise and improve image consistency in 3D UVI.
• To enhance robustness against domain shifts and inter-slice intensity variations.

Main Methods:

• Proposed RFDNet integrating a Deep Frequency Filtering (DFF) module into a standard denoising model.
• The DFF module adaptively filters frequency components in the encoder to suppress noise and balance spectral content.
• Evaluated performance using Doppler phantom, carotid artery, and abdominal datasets.

Main Results:

• RFDNet significantly outperformed conventional methods in peak signal-to-noise ratio (PSNR), structural similarity (SSIM), and root mean squared error (RMSE).
• 2D frequency spectrum analysis confirmed the DFF module's ability to dynamically adjust frequency components and maintain spectral balance.
• Spectral KL divergence analysis demonstrated robustness against slice-wise normalization inconsistencies.

Conclusions:

• RFDNet effectively reduces noise artifacts and improves imaging consistency in 3D UVI.
• The adaptive frequency filtering enhances domain generalization and clinical applicability by improving imaging reliability.
• Future work includes exploring 3D training and architectural refinements for improved computational efficiency.