高级检索

深度学习图像重建联合器官剂量调制在颅脑低剂量CT中的图像质量与晶状体防护优化研究

Deep learning image reconstruction combined with organ dose modulation for optimization in low-dose cranial CT: image quality and lens protection

    摘要
  • 摘要:
    目的 探讨深度学习图像重建（DLIR）算法在补偿器官剂量调制（ODM）技术所致噪声升高方面的有效性，并与传统自适应统计迭代重建-Veo（ASIR-V）算法比较，以实现在低辐射剂量下对颅脑CT图像噪声的优化及灰白质对比度的增强。
    方法 回顾性纳入2022年3月至2023年4月在中山大学附属第三医院接受2次非增强颅脑CT扫描的30例患者，收集其首次检查（A组，常规扫描方案）及短期随访检查（B组，启用ODM的扫描方案）的数据，分别重建2组图像，其中A组采用30% ASIR-V（A-AV30组）；B组采用30% ASIR-V（B-AV30组）、中等强度DLIR（B-DM组）及高等强度DLIR（B-DH组）。记录晶状体区域前、后、左、右方向的管电流，测量CT值、噪声标准差（SD），计算信噪比（SNR）、对比噪声比（CNR）和灰白质比（GWR）。客观测量一致性使用组内相关系数（ICC）评价，主观评价由2名放射科医师采用5分制对图像质量进行评估。
    结果 B组的容积CT剂量指数（CTDIvol）和剂量长度乘积（DLP）均低于A组（均P ＜ 0.05）。客观数据中，各结构的ICC值均超过0.75。在相同30% ASIR-V算法下，B-AV30组的噪声SD值高于A-AV30组，而其SNR、CNR和GWR值均较低（均P ＜ 0.001）。在B组中，采用高等强度DLIR算法（B-DH组）重建图像的噪声SD值最低，SNR、CNR、GWR值及主观评分最高（均P ＜ 0.001）。2名放射科医师的主观评价Kappa值介于0.792 ～ 0.852（均P ＜ 0.001）。
    结论 高级DLIR算法能够有效补偿因ODM技术造成的图像质量下降，在降低辐射剂量、保护晶状体的同时，实现更优的噪声控制和灰白质对比度，其性能优于传统ASIR-V算法，为临床颅脑低剂量CT扫描提供了一种有效的综合优化策略。

     

    Abstract:
    Objective  To evaluate the efficacy of deep learning image reconstruction (DLIR) in compensating for noise elevation induced by organ dose modulation (ODM) technology, and to compare its performance with traditional adaptive statistical iterative reconstruction-Veo (ASIR-V) for optimizing image noise and enhancing gray-white matter contrast in low-dose cranial computed tomography (CT).
    Methods In this retrospective study, 30 patients who underwent two non-contrast cranial CT scans. Data from initial examinations (Group A, conventional protocol) and short-term follow-up examinations (Group B, ODM-enabled protocol) were collected. Images were reconstructed as follows: in Group A, 30% ASIR-V (A-AV30 subgroup) was adopted; in Group B, 30% ASIR-V (B-AV30 subgroup), medium-level DLIR (B-DM subgroup), and high-level DLIR (B-DH subgroup) were employed. Tube current in the anterior, posterior, left, and right directions of the lens region was recorded. CT values and noise (standard deviation, SD) were measured to calculate signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and gray-white matter ratio (GWR). The consistency of objective measurement was evaluated by intra-group correlation coefficient (ICC). Subjective evaluation was performed by two radiologists using a 5-point scale.
    Results Volume CT dose index (CTDIvol) and dose-length product (DLP) in Group B were significantly lower compared to those Group A (both P ＜ 0.05). In the objective measurement data, the ICC values of all structures exceeded 0.75. Under the same 30% ASIR-V algorithm, the B-AV30 subgroup demonstrated higher noise (SD) and lower SNR, CNR, and GWR compared to the A-AV30 subgroup (all P ＜ 0.001). Within Group B, images reconstructed with high-level DLIR (B-DH) exhibited the lowest noise and highest SNR, CNR, GWR, and subjective scores (all P ＜ 0.001). The subjective evaluation Kappa values between two radiologists were ranged from 0.792 to 0.852 (all P ＜ 0.001).
    Conclusions High-level DLIR (B-DH) effectively compensates for image quality degradation caused by ODM technology, achieving superior noise control and gray-white matter contrast while significantly reducing radiation dose and providing lens protection. Its performance surpasses that of traditional ASIR-V algorithm, offering an effective comprehensive optimization strategy for low-dose cranial CT scanning in clinical practice.

     

    • HTML全文
    • 参考文献(31)
    • 相关文章
    • 施引文献
  • 资源附件(0)

/

返回文章
返回