Purpose: To evaluate the impact of super-resolution deep learning reconstruction with 1024 matrix for coronary plaque quantification in coronary CT angiography (CCTA).

Materials and Methods: Twenty patients who clinically diagnosed coronary disease and underwent CCTA in 320-rows CT (Aquilion One Insight Edition) were retrospectively collected. All images were reconstructed with four different reconstruction algorithms: hybrid iterative reconstruction (HIR), deep learning reconstruction (DLR), super-resolution deep learning reconstruction in 512×512 matrix (SR-DLR512), and super-resolution deep learning reconstruction in 1024×1024 matrix (SR-DLR1024). Subjective image quality of plaques was scored using a 5-point Likert scale by two experienced radiologists in four groups. Coronary plaque quantification including plaque area, plaque burden and plaque detection rates were compared among four groups with intravascular ultrasound (IVUS) as the reference standard.

Results: SR-DLR1024 significantly enhanced subjective image quality of plaques compared to HIR, DLR and SR-DLR512 (all P<0.05). SR-DLR significantly improved the plaque detection rates compared to HIR and DLR (all P<0.05), particularly in detection of calcified plaques. Compared to HIR, DLR and SR-DLR512, SR-DLR1024 exhibited the strongest correlation with IVUS in assessing plaque area (r=0.984) and plaque burden (r=0.986).

Conclusion: Compared with conventional reconstruction algorithms, SR-DLR1024 improves subjective image quality of plaques and the accuracy of coronary plaque quantification.