目的 探讨深度学习图像重建（Deep Learning Image Reconstruction，DLIR）对高分辨率CT血管成像图像质量和诊 断准确性的影响。方法 前瞻性纳入56例患者在GE Revolution CT以高分辨率模式进行冠状动脉CT血管成像（Coronary CT Angiography，CCTA），分别使用50%权重的自适应迭代重建技术（Adaptive Statistical Iterative Reconstruction，ASIR)-V和 中级别（DLIR-M）、高级别（DLIR-H）重建原始数据（应用高清卷积核）。记录主动脉根部和主要冠状动脉近段的图像 噪声，并计算信噪比（Signal to Noise Ratio，SNR）和对比噪声比（Contrast to Noise Ratio，CNR）以客观评价图像质量。 在32例患者亚组中，比较ASIR-V 50%、DLIR-M和DLIR-H对冠状动脉狭窄的诊断准确率，并与有创冠状动脉造影进行比 较。主观图像质量由两名有5年以上经验的影像诊断医师按5分制进行评级。结果 与ASIR-V 50%相比，DLIR-H和DLIR-M 的噪声分别降低了54.8%和59.9%（P值均<0.05）丨SNR和CNR显著增加，主观图像质量较ASIR-V 50%有显著改善（P值均 <0.05）。DLIR-H和DLIR-M对冠状动脉狭窄的诊断准确性无显著影响。结论 与ASIR-V相比，DLIR能提升高分辨率模式下 CCTA图像的整体质量，并不影响对冠状动脉阻塞性心脏病的诊断准确性。

Objective To explore the influence of deep learning image reconstruction (DLIR) on image quality and diagnostic accuracy of high-resolution CT angiography. Methods 56 patients underwent coronary computed tomography angiography (CCTA) on a high-definition CT (Revolution CT, GE Healthcare). Data sets were reconstructed with ASiR-V 50% (applying HD kernels) and with DLIR at medium and high settings (DLIR-M and DLIR-H), respectively. The image noise, signal to noise ratio (SNR), and contrast to noise ratio (CNR) on aorta root and main coronary proximal segments were calculated to evaluate image quality objectively. In a subgroup of 32 patients, diagnostic accuracy of ASiR-V 50%, DLIR-M and DLIR-H for diagnosis of coronary artery disease (CAD) were compared with invasive coronary angiography. Subjective image quality was blindly graded by two imaging diagnosticians with over 5 years of experience on a five-point scale. Results The noise of DLIR-M and DLIR-H were significantly decreased by 54.8% and 59.9%, while SNR and CNR were significantly increased compared to ASiR-V 50% (P<0.05). The subjective image quality improved significantly (P<0.05). DLIR-H and DLIR-M had no significant effect on the diagnostic accuracy of coronary artery stenosis. Conclusion Compared with ASIR-V, DLIR could effectively improve the overall image quality of CCTA images in high-resolution mode. and has no effect on diagnostic accuracy of CCTA for CAD detection.