Wavelet-domain frequency-mixing transformer unfolding network for low-dose computed tomography image denoising

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Low-dose computed tomography (LDCT) is crucial for reducing radiation exposure in medical imaging.
• Image noise in LDCT poses a significant challenge, especially in orthopedic imaging where fine bone textures are vital for diagnosis.
• Existing deep learning denoising methods often struggle to balance noise suppression with the preservation of subtle anatomical structures.

Purpose of the Study:

• To develop a novel deep learning model for denoising orthopedic LDCT images.
• To simultaneously address noise reduction and the preservation of fine structural details.
• To overcome the limitations of current spatial-domain denoising techniques.

Main Methods:

• A wavelet-domain frequency-mixing transformer (FMT) network was developed, integrating multiscale wavelet decomposition and FMT blocks for cross-band feature interaction.
• The network incorporates physics-based noise modeling for realistic denoising.
• The model was trained and validated using clinical orthopedic CT scans.

Main Results:

• The proposed FMT network achieved superior denoising performance compared to six state-of-the-art methods.
• Quantitative evaluation showed a mean PSNR of 42.3 dB and an SSIM of 0.961.
• Radiologist evaluation confirmed significantly better preservation of trabecular bone patterns (P<0.01).

Conclusions:

• The developed network represents a new approach to LDCT denoising by leveraging frequency-domain characteristics.
• This method shows particular value for orthopedic imaging, where fine structural fidelity is essential.
• The approach holds potential for further radiation dose reduction in musculoskeletal imaging without compromising diagnostic quality.