CLEAR: Comprehensive Learning Enabled Adversarial Reconstruction for Subtle Structure Enhanced Low-Dose CT Imaging.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• X-ray computed tomography (CT) provides vital anatomical information but poses radiation risks.
• Low-dose CT (LDCT) reduces radiation exposure but introduces noise and artifacts, hindering diagnostic accuracy.
• Existing deep learning (DL) methods for LDCT often cause subtle structure degeneration and blurring.

Purpose of the Study:

• To develop an advanced deep learning method for high-quality low-dose CT reconstruction.
• To address the challenges of noise suppression and structural fidelity in LDCT imaging.
• To improve the visual perception and diagnostic reliability of CT images obtained with reduced radiation doses.

Main Methods:

• Development of the Comprehensive Learning Enabled Adversarial Reconstruction (CLEAR) method.
• Utilizing a generator trained on a comprehensive domain to extract richer features from raw projections.
• Implementing a multi-level loss function and WGAN-GP for enhanced image quality and statistical property migration.

Main Results:

• CLEAR effectively suppresses noise and artifacts in low-dose CT images.
• The method demonstrates superior preservation of subtle anatomical structures compared to existing DL techniques.
• Quantitative and qualitative analyses confirm CLEAR's competitive performance in noise reduction, structural fidelity, and visual perception.

Conclusions:

• CLEAR offers a promising solution for high-quality low-dose CT imaging, balancing radiation dose reduction with diagnostic image quality.
• The CLEAR method mitigates common deep learning artifacts like structure degeneration and blurring.
• This approach has the potential to enhance diagnostic confidence and patient safety in medical CT applications.