Normal-resolution vs. super-resolution deep learning reconstruction for diagnosis of functionally significant coronary stenosis using cardiac CT.

Abstract

<h4>Background</h4>Super-resolution deep learning reconstruction (SR-DLR) has been developed to reduce image noise and enhance spatial resolution beyond that of normal-resolution deep learning reconstruction (NR-DLR).<h4>Purpose</h4>To compare the diagnostic performance of CT-derived fractional flow reserve (CT-FFR) against invasive FFR using NR-DLR and SR-DLR.<h4>Methods</h4>In this single-center retrospective study, 129 patients (mean age, 69 years ​±11 [SD]; 94 men) who underwent coronary CT angiography followed by invasive FFR between February 2022 and March 2025 were included. CT-FFR was computed using a mesh-free simulation model. Functionally significant stenosis was defined as FFR ≤0.80. The diagnostic performance of CT-FFR was compared between NR-DLR and SR-DLR using receiver operating characteristic curve analysis.<h4>Results</h4>The mean invasive FFR was 0.81 ​± ​0.08, and 70 out of 157 vessels (45 ​%) had FFR ≤0.80. The mean signal-to-noise ratio was higher with SR-DLR than with NR-DLR (33.3 ​± ​6.6 vs. 23.9 ​± ​4.5, p ​< ​0.001). The area under the receiver operating characteristic curve for detecting functionally significant stenosis was higher with SR-DLR (0.85; 95 ​% CI: 0.78, 0.91) than with NR-DLR (0.72; 95 ​% CI: 0.64, 0.81; p ​< ​0.001). Diagnostic accuracy was also higher with SR-DLR (85 ​%; 134 out of 157 vessels; 95 ​% CI: 79, 90) than with NR-DLR (74 ​%; 116 out of 157 vessels; 95 ​% CI: 66, 81; p ​< ​0.001).<h4>Conclusions</h4>Compared with NR-DLR, SR-DLR enhances image quality and improves the diagnostic performance of CT-FFR for identifying functionally significant stenosis.