An attention-based deep convolutional neural network for ultra-sparse-view CT reconstruction.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• X-ray Computed Tomography (CT) is crucial for diagnosis but involves radioactivity exposure.
• Sparse-view CT minimizes radiation but introduces streaking artifacts in reconstructed images.

Purpose of the Study:

• To develop an advanced deep learning model for correcting artifacts in sparse-view CT images.
• To enhance the diagnostic quality of CT scans while reducing patient radiation dose.

Main Methods:

• An end-to-end attention-based deep network, integrating U-Net and ResNet50, was designed for artifact correction.
• The network processes images reconstructed using filtered back-projection from sparse sinograms.
• Attention mechanisms were employed to focus on relevant features and suppress irrelevant ones.

Main Results:

• The proposed model effectively removed streaking artifacts and preserved crucial structural details in CT images.
• Quantitative analysis showed significant improvements in Peak Signal-to-Noise Ratio (PSNR), Structural Similarity (SSIM), and Root Mean Squared Error (RMSE).
• Achieved an average PSNR of 33.9538, SSIM of 0.9435, and RMSE of 45.1208 at 20 views, with validated transferability on the AAPM dataset.

Conclusions:

• The attention-based deep network offers a promising solution for high-quality sparse-view CT image reconstruction.
• This approach can potentially reduce radiation risks in medical imaging without compromising diagnostic accuracy.
• The model demonstrates superior performance in artifact reduction and image quality enhancement compared to existing methods.