Super-Resolution Deep Learning Reconstruction for Coronary CT Angiography: Coronary Stenosis Assessment and CAD-RADS...

Background A novel super-resolution deep learning reconstruction (SR-DLR) algorithm, trained using data acquired with ultra-high-resolution CT, can potentially enhance spatial resolution in coronary CT angiography (CCTA), improving stenosis assessment; however, evidence is limited. Purpose To compare the performance of SR-DLR versus hybrid iterative reconstruction (HIR) in assessing coronary stenosis, using invasive coronary angiography (ICA) as the reference standard, and to explore the potential impact on patient-level Coronary Artery Disease Reporting and Data System (CAD-RADS) classification. Materials and Methods From September 2023 to November 2024, patients who underwent clinically indicated CCTA and ICA within a 2-month interval were prospectively enrolled at 10 hospitals across China. CCTA images were reconstructed with both HIR and SR-DLR, and percentage diameter stenosis (PDS) of calcified, noncalcified, and mixed plaques was quantified. Participant-level CAD-RADS category was determined based on the highest-grade stenosis. Using ICA as the reference standard, diagnostic performance of HIR and SR-DLR in detecting significant stenosis (50% or greater stenosis) was compared using the area under the receiver operating characteristic curve (AUC). Results The study included 204 individuals (mean age, 64.3 years ± 9.1 [SD]; 137 male participants) with 605 plaques (175 calcified, 140 noncalcified, 290 mixed). Median PDS for calcified plaques was lower with SR-DLR than with HIR (58% [IQR, 44%-71%] vs 63% [IQR, 53%-85%]; P < .001), with no evidence of a difference in median PDS for noncalcified (P = .09) or mixed (P = .40) plaques. Forty-one individuals were assigned a different CAD-RADS category with SL-DLR relative to HIR: 25 downgraded and 16 upgraded. SR-DLR outperformed HIR in detecting significant stenosis at the lesion level (AUC, 0.97 [95% CI: 0.96, 0.98] vs 0.90 [95% CI: 0.87, 0.92]; P < .001) and participant level (AUC, 0.90 [95% CI: 0.82, 0.98] vs 0.79 [95% CI: 0.70, 0.89]; P < .001). Conclusion SR-DLR outperformed HIR for coronary stenosis assessment and led to 20% (41 of 204) participant-level CAD-RADS reclassification. © The Author(s) 2026. Published by the Radiological Society of North America under a CC BY 4.0 license. Chinese Clinical Trial Registry no. ChiCTR2300075364 Supplemental material is available for this article.