Deep learning-based denoising in cardiac CT: effects on image quality, calcium scoring interchangeability, and...

Area of Science:

• Cardiology
• Radiology
• Medical Imaging

Background:

• Ischemic heart disease poses a global health challenge, necessitating accurate and timely diagnosis.
• Coronary artery calcium (CAC) scoring and coronary computed tomography angiography (CCTA) are vital diagnostic tools, but image noise can impede assessment.
• Deep learning-based denoising (DLD) algorithms offer potential for enhancing cardiac CT image quality, yet their clinical impact remains under investigation.

Purpose of the Study:

• To evaluate the impact of deep learning-based denoising (DLD) on cardiac CT image quality, clinical interchangeability, and workflow efficiency.
• To compare DLD with traditional iterative reconstruction (IR) in the context of CAC scoring and CCTA.
• To assess DLD's effect on objective image metrics and diagnostic workflow.

Main Methods:

• Retrospective analysis of 100 patients with paired CAC and CCTA scans.
• Generation of 400 datasets using both IR and DLD reconstruction methods.
• Radiological assessment using semiquantitative scoring, objective metrics (noise, CNR), and workflow efficiency evaluation.

Main Results:

• DLD demonstrated superior image quality compared to IR (p < 0.001), with improved noise and contrast-to-noise ratio (CNR).
• While initial Agatston scores differed (IR higher), manual correction rendered scores clinically comparable between IR and DLD (p ≥ 0.158).
• DLD significantly reduced manual correction time (p < 0.001), indicating enhanced workflow efficiency.

Conclusions:

• The DLD algorithm enhances cardiac CT image quality and streamlines radiological workflows.
• DLD shows potential for improving diagnostic accuracy and patient care in the assessment of ischemic heart disease.
• Further prospective studies are recommended to validate these findings with more objective measures.