深度学习图像重建用于80-kVp胰腺CT协议:与混合迭代重建的图像质量和胰腺导管腺癌可见性比较
该回顾性研究纳入56例因胰腺疾病评估接受80-kVp胰腺协议CT的患者,其中观察到20例胰腺导管腺癌(PDAC),比较混合迭代重建(40%自适应统计迭代重建-Veo)与深度学习图像重建中等强度和高强度水平的图像质量及PDAC可见性。结果显示高强度深度学习重建组背景噪声更低、信噪比和肿瘤-胰腺对比噪声比更高,图像噪声、整体图像质量和PDAC可见性均优于其他两组。
要点
- 高强度深度学习重建(DLIR-H)较混合迭代重建和中等强度DLIR降低背景噪声、提高信噪比和肿瘤-胰腺CNR。
- DLIR-H组的图像噪声、整体图像质量和PDAC可见性主观评分均优于其他两组。
- 80-kVp胰腺协议CT中,高强度深度学习重建可改善图像质量和PDAC可见性。
核心结论
- 在80-kVp胰腺CT协议中,高强度水平深度学习图像重建改善了图像质量和PDAC可见性。
研究设计与样本
- 回顾性研究,纳入2022年1月至2022年7月接受80-kVp胰腺协议CT的56例患者,其中观察到20例PDAC;CT原始数据分别用40%自适应统计迭代重建-Veo(混合-IR组)和深度学习图像重建中等强度及高强度水平(DLIR-M组和DLIR-H组)重建。
关键影像表现/征象
- 除胰腺外,腹主动脉、胰腺和PDAC在胰腺期的CT衰减值以及门静脉和肝脏在门静脉期的CT衰减值在三组间相当(P=.26-.86)。
- DLIR-H组背景噪声低于其他两组(P<.001)。
- DLIR-H组各解剖结构的信噪比高于其他两组(P<.001)。
- DLIR-H组肿瘤-胰腺对比噪声比高于其他两组(P<.001)。
诊断标准与关键数值
- 图像噪声、整体图像质量和PDAC可见性的置信度评分采用五点量表定性评定。
- DLIR-H组图像噪声、整体图像质量和PDAC可见性评分优于其他两组(P<.001-.003)。
临床应用与阅片要点
- 在80-kVp胰腺协议CT中,高强度深度学习图像重建可改善图像质量和PDAC可见性。
局限性
- 原文未报告局限性。
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
PURPOSE: To evaluate the image quality and visibility of pancreatic ductal adenocarcinoma (PDAC) in 80-kVp pancreatic CT protocol and compare them between hybrid-iterative reconstruction (IR) and deep-learning image reconstruction (DLIR) algorithms.
METHOD: A total of 56 patients who underwent 80-kVp pancreatic protocol CT for pancreatic disease evaluation from January 2022 to July 2022 were included in this retrospective study. Among them, 20 PDACs were observed. The CT raw data were reconstructed using 40% adaptive statistical IR-Veo (hybrid-IR group) and DLIR at medium- and high-strength levels (DLIR-M and DLIR-H groups, respectively). The CT attenuation of the abdominal aorta, pancreas, and PDAC (if present) at the pancreatic phase and those of the portal vein and liver at the portal venous phase; background noise; signal-to-noise ratio (SNR) of these anatomical structures; and tumor-to-pancreas contrast-to-noise ratio (CNR) were calculated. The confidence scores for the image noise, overall image quality, and visibility of PDAC were qualitatively assigned using a five-point scale. Quantitative and qualitative parameters were compared among the three groups using Friedman test.
RESULTS: The CT attenuation of all anatomical structures were comparable among the three groups (P = .26-.86), except that of the pancreas (P = .001). Background noise was lower (P <.001) and SNRs (P <.001) and tumor-to-pancreas CNR (P <.001) were higher in the DLIR-H group than those in the other two groups. The image noise, overall image quality, and visibility of PDAC were better in the DLIR-H group than in the other two groups (P <.001-.003).
CONCLUSION: In 80-kVp pancreatic CT protocol, DLIR at a high-strength level improved image quality and visibility of PDAC.
引用
- Takai Y, Noda Y, Asano M, et al. Deep-learning image reconstruction for 80-kVp pancreatic CT protocol: Comparison of image quality and pancreatic ductal adenocarcinoma visibility with hybrid-iterative reconstruction. Eur J Radiol. 2023. PMID: 37423016. doi:10.1016/j.ejrad.2023.110960
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。