Deep-learning Image Reconstruction Algorithm for Improving Coronary Artery Imaging Quality in Pediatric High Heartrate

To evaluate the effectiveness of deep-learning image reconstruction (DLIR) algorithm in improving the image quality of cardiovascular computed tomography angiography (CCTA) for children with high heartrate, this study enrolled 27 consecutive patients (mean age: 9.3±3.1 years) with cardiac symptoms, suspected coronary artery stenosis, and abnormal origins with high heartrates (mean heartrate: 83.7±17.5 bpm). The participants underwent CCTA using a 256-row detector CT scanner under a prospective ECG-triggered single-beat protocol. Images were reconstructed with Adaptive Statistical Iterative Reconstruction V (50% ASIR-V, 80% ASIR-V), and Deep Learning Image Reconstruction–High (DLIR-H). On a 5-point scale, the image quality was assessed independently by two senior radiologists in a blinded manner. The subjective DLIR-H scores were significantly higher than those of the 50% ASIR-V (\chi^2 = 7.19) and 80% ASIR-V (\chi^2 = 6.52) images. On objective measurements, the differences in paracardiac fat noise values among the three groups was statistically significant, and the DLIR-H group had the lowest noise. Compared with the 50% ASIR-V group, the DLIR-H group had significantly higher CNR values on CCTA images of the aortic root (F = 5.84), right coronary artery (F = 4.30), left trunk (F = 3.95), left anterior descending (F = 4.16, P = 0.02), and left gyratory branch (F = 4.29). Thus, in pediatric patients with cardiac symptoms and high heartrates, the deep-learning reconstruction algorithm significantly improved the CCTA image quality, reduced image noise, and improved the image contrast-to-noise ratio and the CCTA diagnostic rate.