Sinogram-characteristic-informed network for efficient restoration of low-dose SPECT projection data.

Area of Science:

• Medical Imaging
• Computational Imaging
• Radiology

Background:

• Low-dose Single Photon Emission Computed Tomography (SPECT) imaging presents significant restoration challenges.
• Existing methods often fail to adequately utilize the inherent characteristics of sinograms for effective restoration.
• There is a critical need for innovative approaches to improve low-dose SPECT sinogram quality.

Purpose of the Study:

• To introduce the Sinogram-characteristic-informed network (SCI-Net) for low-dose SPECT sinogram restoration.
• To develop a model that learns from sinogram properties like continuity, periodicity, and multi-scale line features.
• To enhance the model's understanding and performance in the sinogram restoration process.

Main Methods:

• SCI-Net employs novel mechanisms, including a channel attention module with decay for continuity and a position attention module for global correlations.
• A multi-stage progressive integration mechanism balances local details and overall structure.
• Customized regularization terms tailored to sinogram characteristics are integrated into the loss function for training.

Main Results:

• SCI-Net demonstrated superior performance in both simulated and clinical evaluations compared to existing methods.
• For simulated data, peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) improved significantly on both sinograms and reconstructed images.
• Clinical evaluation showed SCI-Net effectively reduced coefficient of variation (COV) in regions of interest, enhancing reconstructed image quality.

Conclusions:

• The proposed SCI-Net shows promising results for low-dose SPECT projection data restoration.
• Novel mechanisms within SCI-Net effectively leverage sinogram characteristics for improved restoration outcomes.
• SCI-Net offers a valuable tool for enhancing the quality of low-dose SPECT imaging.