深度学习降噪对超高分辨率光子计数CT神经血管成像的影响
该单中心回顾性研究纳入42例接受头颈超高分辨率光子计数探测器CT血管成像的患者,评估事后人工智能降噪算法对图像质量与诊断信心的影响。结果显示降噪显著改善客观指标如SNR与CNR,但主观图像质量与诊断信心无显著改善。
要点
- UHR PCD-CTA在下颈部及易产生伪影区域噪声仍是主要限制因素。
- 事后AID可显著降低各血管层面噪声并改善SNR和CNR。
- 主观图像质量与诊断信心未显著提升,提示高质量区域存在天花板效应。
核心结论
- 事后AID显著改善神经血管UHR PCD-CTA的客观图像质量指标,但主观图像质量和诊断信心无显著变化。
研究设计与样本
- 回顾性单中心研究,纳入42例接受临床适应证头颈UHR PCD-CTA检查的患者。
- 图像采用高级别供应商特定迭代重建(4级)重建,并评估叠加事后AID前后的差异。
关键影像表现/征象
- 下颈部绝对噪声降低幅度最大,各血管层面噪声均显著降低。
- 血管锐度存在统计学显著但绝对差异较小,FWHM测量保持稳定提示空间分辨率未受损。
诊断标准与关键数值
- 定量分析包括下颈部、中颈部及颅内血管层面的SNR、CNR和血管锐度。
- 主观图像质量与诊断信心由两名神经放射科医师独立评估,各血管层面均无显著差异。
临床应用与阅片要点
- 在日常阅片中可考虑将事后AID用于改善下颈部等高噪声区域的客观图像质量。
- 由于主观诊断信心未见显著提升,报告时不应因使用AID而降低对原始图像质量的关注。
局限性
- 单中心回顾性研究设计,样本量有限。
- 主观评分未改善可能受天花板效应影响,难以反映算法在较差图像区域的潜在价值。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
英文原文(摘要)· 点击展开
BACKGROUND AND PURPOSE: Ultrahigh-resolution (UHR) photon-counting detector (PCD)-CT angiography enables detailed visualization of the neurovasculature. However, image noise remains a limiting factor, particularly in the lower neck and artifact-prone regions. We evaluated the impact of a post hoc artificial intelligence-based denoising (AID) algorithm on image quality and diagnostic confidence in neurovascular UHR PCD-CTA.
MATERIAL AND METHODS: This retrospective, single-center study included 42 patients who underwent clinically indicated head and neck UHR PCD-CTA. Images were reconstructed using high-level vendor-specific iterative reconstruction (level 4) with and without additional post hoc AID. Quantitative analyses included signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and vessel sharpness measurements at the lower neck, mid neck, and intracranial vasculature. Subjective image quality and diagnostic confidence were independently assessed by two neuroradiologists.
RESULTS: AID significantly reduced image noise across all vascular levels (p < 0.001), with the largest absolute reduction in the lower neck. Vessel sharpness demonstrated statistically significant but small absolute differences, with preserved spatial resolution as reflected by stable FWHM measurements. No significant differences in subjective image quality and diagnostic confidence were observed at any vascular level.
CONCLUSIONS: Post hoc AID substantially improves objective image quality in neurovascular UHR PCD-CTA by improving objective image quality metrics. The lack of improvement in subjective assessments suggests a ceiling effect in regions with already high image quality and highlights persistent challenges in the lower neck and artifact prone regions.
引用
- Tóth, Hagar, Silbergleit, et al. Impact of Deep Learning-Based Denoising on Image Quality and Diagnostic Confidence in Neurovascular Ultrahigh-Resolution Photon-Counting CT Angiography. AJNR Am J Neuroradiol. 2026. PMID: 42386370. doi:10.3174/ajnr.A9506
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。