Super-Resolution Deep Learning Reconstruction for Improved Image Quality of Coronary CT Angiography

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Coronary CT angiography (CCTA) is crucial for diagnosing coronary artery disease.
• Image quality, including noise and edge sharpness, impacts diagnostic accuracy.
• Deep learning reconstruction (DLR) techniques offer potential for improving CCTA image quality.

Purpose of the Study:

• To compare image noise and edge sharpness between super-resolution DLR (SR-DLR) and conventional DLR (C-DLR) in CCTA.
• To evaluate the agreement in coronary stenosis grading between CCTA (using SR-DLR and C-DLR) and invasive coronary angiography (ICA).

Main Methods:

• Retrospective analysis of CCTA data from 58 patients (320-row CT).
• Images were reconstructed using both SR-DLR and C-DLR algorithms.
• Quantitative assessment of image noise, signal-to-noise ratio, edge sharpness, and stent FWHM.
• Comparison of stenosis grading agreement with ICA using weighted kappa statistics.

Main Results:

• SR-DLR significantly reduced image noise by 31% compared to C-DLR (12.6 HU vs 18.2 HU).
• SR-DLR demonstrated significant improvements in signal-to-noise ratio and edge sharpness.
• The full width at half maximum (FWHM) of stents was significantly thinner with SR-DLR (0.72 mm vs 1.01 mm).
• Agreement in stenosis grading between CCTA and ICA was higher with SR-DLR (κ = 0.83) than with C-DLR (κ = 0.77).

Conclusions:

• SR-DLR enhances vessel sharpness and reduces image noise in CCTA.
• SR-DLR improves the accuracy of coronary artery stenosis grading compared to C-DLR.
• SR-DLR represents a promising advancement for CCTA image reconstruction.