基于AI赋能的低剂量造影剂对冠脉病变的诊断进展

Progress in AI-Enabled Low-Dose Contrast Agent for the Diagnosis of Coronary Artery Lesions

摘要: 冠状动脉CT血管成像(CCTA)是冠心病无创评估的重要工具，但常规检查依赖较高碘负荷和较快注射流速，限制了其在肾功能边缘受损、静脉条件较差及需重复随访患者中的应用。近年来，低剂量造影剂冠脉成像已从单纯依赖扫描参数优化，逐步发展为“采集端增益–重建端补偿–分析端定量”的协同模式。现有研究表明，低keV虚拟单能图像可提高碘信号利用效率，在部分研究中实现约50%的碘剂量下降而仍保持可诊断图像质量；深度学习重建(DLR/DLIR)可在低体积、低浓度甚至低流速方案下明显降低噪声、提升信噪比和主观评分；超分辨率重建、CE-boost及光子计数CT (PCD-CT)进一步推动“多低”方案走向临床可行。与此同时，AI-QCT、自动斑块定量、高危斑块识别及CT-FFR相关模型，正在使低剂量成像从“可读”走向“可量化、可分层、可支持决策”。总体看，AI显著拓展了低剂量造影剂CCTA的应用边界，但其稳定推广仍需依赖多中心外部验证、复杂病变亚组研究及临床终点证据。

Abstract: Coronary computed tomography angiography (CCTA) is an important noninvasive tool for the evaluation of coronary artery disease. However, conventional CCTA generally relies on a relatively high iodine load and rapid injection rates, which limit its use in patients with borderline renal impairment, poor venous access, or the need for repeated follow-up examinations. In recent years, low-contrast-dose coronary imaging has evolved from a strategy that primarily depended on scan-parameter optimization to a collaborative paradigm integrating acquisition-side enhancement, reconstruction-side compensation, and analysis-side quantification. Existing studies have shown that low-keV virtual monoenergetic imaging can improve the efficiency of iodine signal utilization, enabling approximately a 50% reduction in iodine dose in some studies while still maintaining diagnostic image quality. Deep learning reconstruction (DLR/DLIR) can substantially reduce image noise and improve signal-to-noise ratio as well as subjective image quality scores under low-volume, low-concentration, and even low-injection-rate protocols. In addition, super-resolution reconstruction, contrast enhancement boost (CE-boost), and photon-counting CT (PCD-CT) have further advanced the clinical feasibility of “multi-low” protocols. Meanwhile, AI-based quantitative CT (AI-QCT), automated plaque quantification, high-risk plaque identification, and CT-derived fractional flow reserve (CT-FFR)-related models are driving low-contrast-dose imaging beyond mere interpretability toward quantification, risk stratification, and decision support. Overall, AI has markedly expanded the application boundaries of low-contrast-dose CCTA; however, its stable and widespread implementation still depends on multicenter external validation, subgroup studies of complex lesions, and evidence from clinical endpoints.