ISSN 1004-4140
    CN 11-3017/P

    超分辨率深度学习重建算法在冠状动脉CT血管成像图像质量优化中的应用：一项Meta分析与系统评价

    Application of Super-Resolution Deep Learning-based Reconstruction in Coronary CT Angiography Image Quality Optimization: A Meta-Analysis and Systematic Review

      摘要
    • 摘要: 目的：探讨超分辨率深度学习重建（SR-DLR）在冠状动脉CT血管成像（CCTA）中对图像质量优化效能。方法：通过计算机检索PubMed、Cochrane Library、Web of Science、Embase、维普、万方、知网等数据库，检索时限为各数据库建库至 2025年3月1日，筛选对比SR-DLR与其他重建算法应用于CCTA的研究。依据纳入及排除标准，采用美国卫生保健研究和质量机构（AHRQ）推出的横断面研究偏倚风险评价标准对文献进行质量评估，提取文献中使用的评估标准及相关数据，并对相应数据进行合并。采用Review Manager 5.3和Stata 17.0软件进行数据分析，包括异质性检验和效应量计算。结果：Meta分析结果显示，与传统的混合迭代重建（HIR）、基于模型的迭代重建（MBIR）和正常分辨率深度学习重建（NR-DLR）相比，SR-DLR在降低图像噪声，提高对比噪声比（CNR）、信噪比（SNR）、提高整体图像质量方面具有显著优势。具体而言，SR-DLR显著降低图像噪声：SMDSR-DLR vs. HIR=−2.64（95%CI −3.40~−1.88，P＜0.01），SMDSR-DLR vs. MBIR=−2.90（95%CI −4.82~−0.98，P＜0.01），SMDSR-DLR vs. NR-DLR=−1.42（95%CI −2.06~−0.77，P＜0.01）。在CNR方面，SR-DLR也显著升高：SMDSR-DLR vs. HIR=1.85（95%CI 1.34~2.36，P＜0.01），SMDSR-DLR vs. MBIR=1.43（95%CI 0.37~2.49，P=0.01），SMDSR-DLR vs. NR-DLR=0.72（95%CI 0.49~0.94，P＜0.01）。在SNR方面，SR-DLR显著高于HIR、MBIR和NR-DLR。SMDSR-DLR vs. HIR=1.23（95%CI 0.91~1.54，P＜0.01），SMDSR-DLR vs. MBIR=0.38（95%CI：0.05~0.72，P=0.02），SMDSR-DLR vs. NR-DLR=0.45（95%CI 0.24~0.65，P＜0.01）。此外，SR-DLR在整体图像质量方面显著优于其他算法：SMDSR-DLR vs. HIR=9.07（95%CI 5.48~12.66，P＜0.01），SMDSR-DLR vs. MBIR=1.89（95%CI 1.58~2.21，P＜0.01），SMDSR-DLR vs. NR-DLR=1.90（95%CI 0.48~3.32，P=0.01）。结论：SR-DLR技术优化冠状动脉CT血管成像的图像质量，具有广阔的临床应用前景。

       

      Abstract: Objective: To systematically review and meta-analyze the effectiveness of super-resolution deep learning-based reconstruction (SR-DLR) in optimizing the quality of coronary computed tomography angiography (CCTA) images. Methods: The PubMed, Cochrane Library, Web of Science, Embase, VIP, Wanfang, and CNKI databases were searched to identify studies that compared SR-DLR with other reconstruction algorithms applied to CCTA. Studies reported from the date of database inception to March 1, 2025 were selected on the basis of inclusion and exclusion criteria. The quality of the literature was assessed using the Agency for Healthcare Research and Quality (AHRQ) risk of bias assessment criteria for cross-sectional studies. Evaluation criteria and relevant data were extracted from the literature, and the corresponding data were pooled. Data analyses, including heterogeneity testing and effect size calculation, were performed using Review Manager 5.3 and Stata 17.0 software. Results: According to the meta-analysis results, SR-DLR showed significant advantages over traditional hybrid iterative reconstruction (HIR), model-based iterative reconstruction (MBIR), and normal-resolution deep learning-based reconstruction (NR-DLR) in reducing image noise, improving both the contrast-to-noise ratio (CNR) and signal-to-noise ratio (SNR), and enhancing overall image quality. Specifically, SR-DLR significantly reduced image noise: standard mean difference (SMD)SR-DLR vs. HIR=−2.64 (95%CI −3.40~−1.88, P<0.01), SMDSR-DLR vs. MBIR=−2.90 (95%CI −4.82~−0.98, P<0.01), and SMDSR-DLR vs. NR-DLR=−1.42 (95%CI −2.06~−0.77, P<0.01). SR-DLR also significantly improved the CNR: SMDSR-DLR vs. HIR=1.85 (95%CI 1.34~2.36, P<0.01), SMDSR-DLR vs. MBIR=1.43 (95%CI 0.37~2.49, P=0.01), and SMDSR-DLR vs. NR-DLR=0.72 (95%CI 0.49~0.94, P<0.01). With regard to SNR improvement, SR-DLR was significantly superior to HIR, MBIR, and NR-DLR: SMDSR-DLR vs. HIR=1.23 (95% CI 0.91~1.54, P<0.01), SMDSR-DLR vs. MBIR=0.38 (95% CI 0.05~0.72, P=0.02), and SMDSR-DLR vs. NR-DLR=0.45 (95% CI 0.24~0.65, P<0.01). Furthermore, SR-DLR significantly outperformed the other algorithms in enhancing overall image quality: SMDSR-DLR vs. HIR=9.07 (95% CI 5.48~12.66, P<0.01), SMDSR-DLR vs. MBIR=1.89 (95% CI 1.58~2.21, P<0.01), and SMDSR-DLR vs. NR-DLR=1.90 (95% CI 0.48~3.32, P=0.01). Conclusion: SR-DLR was concluded to be the best technology for optimizing the image quality of CCTA and holds broad prospects for clinical application.

       

      • HTML全文
      • 参考文献(18)
      • 相关文章
      • 施引文献
    • 资源附件(0)

    /

    返回文章
    返回