ISSN 1004-4140
    CN 11-3017/P

    深度学习重建算法在腹部多剂量CT图像质量优化与辐射剂量控制的体模研究

    CT Image Quality Optimization and Radiation Dose Control Using Deep Learning Reconstruction Algorithms in Multi-Dose Abdominal CT: A Phantom Study

      摘要
    • 摘要: 目的：探讨深度学习重建算法（DLIR）相较于自适应统计迭代重建算法（ASiR-V）和滤波反投影法（FBP）在腹部CT图像质量优化和辐射剂量控制中的价值。方法：采用256排CT对头颈躯干类人体模（CTU-41）进行多剂量（5、10、15和20 mGy）腹部CT扫描，每个剂量水平下均进行3次重复扫描。原始图像经过FBP、ASiR-V（30%、50%、70%）及DLIR（L、M、H）重建生成28组图像。由两名放射医师对图像进行主客观评价，客观评价包括信噪比（SNR）、对比度噪声比（CNR）、图像噪声（N）、噪声等效剂量指数（NED）、图像质量剂量指数（IQF）；主观评价采用Likert5分量表法。采用析因设计的方差分析比较不同辐射剂量、不同算法及水平重建图像的客观评价指标，采用多因素方差分析比较各组的主观评价。结果：在相同剂量下，DLIR-H图像的SNR、CNR和NED显著高于其他算法，N、IQF显著低于其他算法。随剂量增高，SNR、CNR和NED呈阶梯式升高，噪声、IQF逐渐降低。高剂量（20 mGy）下，DLIR-H的肝脏SNR较ASiR-V50%提升55.9%，CNR提升61.2%。主观评价显示，DLIR-H和DLIR-M重建图像的主观评价显著优于DLIR-L、ASiR-V70%和ASiR-V50%，观察者一致性较高（ICC=0.933/0.893）。腹部CT以20 mGy，ASiR-V50%为参照组，DLIR-H（10/15/20 mGy），DLIR-M（15/20 mGy）和ASiR-V70%（15/20 mGy）的客观指标、主观评价指标均优于参考组。DLIR算法在10 mGy剂量下实现与参考组相当的图像质量，辐射剂量减少50%。结论：DLIR-H可显著提升腹部CT图像质量并降低辐射剂量，其在高剂量下噪声抑制能力突出，低剂量下仍可满足诊断需求，为临床优化辐射剂量方案提供了可靠技术支撑。

       

      Abstract:
      Objective To compare the value of a deep learning image reconstruction (DLIR) algorithm with those of adaptive statistical iterative reconstruction V (ASiR-V) and filtered back projection (FBP) algorithms in optimizing abdominal CT image quality and controlling radiation dose.
      Methods A 256-slice CT scanner was used to perform abdominal CT scans on a head-neck-trunk phantom (CTU-41) at four predefined radiation dose levels (5, 10, 15, and 20 mGy CTDIvol). Three repeated scans were conducted at each dose level to ensure reproducibility. Raw images were reconstructed using FBP, ASiR-V (30%, 50%, and 70%), and DLIR (L, M, and H), resulting in 28 groups of images. Subsequently, two radiologists performed objective and subjective evaluations. Objective evaluations included signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), image noise (N), noise equivalent dose index (NED), and image quality figure (IQF). Subjective evaluation was conducted using a 5-point Likert scale. Factorial analysis of variance (ANOVA) was used to compare objective evaluation indices of reconstructed images with different radiation dose levels and algorithms, whereas multivariate ANOVA was applied to compare subjective evaluations across groups.
      Results At the same dose level, DLIR-H images showed significantly higher SNR, CNR, and NED but significantly lower N and IQF than other algorithms. SNR, CNR, and NED demonstrated stepwise increases with increasing dose, while N and IQF gradually decreased. At high-dose level (20 mGy), DLIR-H achieved a 55.9% increase in SNR and 61.2% increase in CNR in the liver compared with ASiR-V50%. Subjective evaluation showed that DLIR-H and DLIR-M reconstructed images were significantly superior to those of DLIR-L, ASiR-V70%, and ASiR-V50%, with high inter-observer agreement (ICC=0.933/0.893). Using 20 mGy with ASiR-V50% as the reference group in abdominal CT, DLIR-H (10/15/20mGy), DLIR-M (15/20mGy), and ASiR-V70% (15/20mGy) outperformed the reference group in objective and subjective evaluation indices. The DLIR algorithm achieved image quality similar to that of the reference group at 10mGy, reducing the radiation dose by 50%.
      Conclusions DLIR-H significantly improved abdominal CT image quality and reduced radiation dose. It demonstrated outstanding noise suppression at high-dose levels and still met diagnostic requirements at low-dose levels. This provides reliable technical support for clinical optimization of radiation dose protocols.

       

      • HTML全文
      • 参考文献(34)
      • 相关文章
      • 施引文献
    • 资源附件(0)

    /

    返回文章
    返回