Influence of deep learning-based super-resolution reconstruction on Agatston score

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Cardiovascular Imaging

Background:

• Deep learning-based super-resolution reconstruction (DLSRR) is increasingly used in medical imaging.
• Its impact on quantitative metrics like the Agatston score in cardiac CT is not fully understood.
• Evaluating DLSRR's effect on image quality and coronary artery calcium (CAC) assessment is crucial for clinical application.

Purpose of the Study:

• To assess the influence of DLSRR on cardiac CT image quality.
• To determine the effect of DLSRR on Agatston scores and coronary artery calcium (CAC) volume.
• To evaluate the concordance of CAC risk classification between DLSRR and traditional filtered back projection (FBP).

Main Methods:

• Cardiac CT datasets from 111 patients were reconstructed using filtered back projection (FBP) and three DLSRR strengths.
• Image quality was evaluated by measuring noise, SNR, CNR, and edge rise slope (ERS).
• Agatston scores, CAC volumes, and Coronary Artery Calcium Data and Reporting System (CAC-DRS) classifications were compared.

Main Results:

• DLSRR significantly reduced image noise and improved SNR and CNR, with enhanced effects at higher strengths.
• Edge rise slope (ERS) was significantly improved by DLSRR compared to FBP, but strengths of DLSRR showed no significant difference.
• Agatston scores and CAC volumes remained unaffected by DLSRR (p > 0.90).
• The concordance rate for CAC-DRS classification between FBP and DLSRR was high at 93%.

Conclusions:

• DLSRR effectively enhances cardiac CT image quality by reducing noise and improving sharpness.
• DLSRR does not significantly alter Agatston scores or CAC volumes, preserving quantitative assessment.
• While concordance is high, DLSRR may lead to some reclassifications in CAC-DRS risk categories, necessitating cautious clinical implementation.