PWD: Prior-Aware Wavelet Diffusion for Efficient Dental Limited-Angle CT Reconstruction.

Area of Science:

• Medical Imaging
• Computational Imaging
• Artificial Intelligence in Radiology

Background:

• Generative diffusion models are increasingly used in medical imaging, especially for limited-angle computed tomography (LACT).
• Standard diffusion models require extensive sampling steps for high-quality reconstruction, leading to computational burdens.
• Existing skip-sampling methods improve efficiency but often sacrifice fine structural details.

Purpose of the Study:

• To develop an efficient diffusion model for dental limited-angle CT reconstruction that preserves fine structural details.
• To mitigate image degradation issues associated with skip-sampling strategies in LACT.
• To enable high-quality reconstruction with significantly fewer sampling steps.

Main Methods:

• Proposed Prior-aware Wavelet Diffusion (PWD) model for efficient LACT reconstruction.
• PWD learns structural correspondences between LACT and fully sampled images during training.
• During inference, LACT images act as priors to guide sampling, and multi-scale wavelet domain fusion enhances fine details.

Main Results:

• PWD demonstrated superior performance over existing methods in dental LACT reconstruction under identical sampling conditions.
• Achieved at least 1.7 dB improvement in Peak Signal-to-Noise Ratio (PSNR) and a 10% gain in Structural Similarity Index Measure (SSIM) using only 50 sampling steps.
• Effectively preserved fine structural details and mitigated degradation typically caused by skip-sampling.

Conclusions:

• PWD offers an efficient and effective solution for dental limited-angle CT reconstruction.
• The proposed method balances reconstruction fidelity and computational efficiency.
• PWD shows significant potential for improving diagnostic accuracy in dental imaging applications.