文献速递神经外周血管CT

深度学习重建算法在双低扫描双能颈动脉CT血管成像中改善40 keV虚拟单能图像质量的比较

Comparison of deep learning reconstruction algorithms to improve image quality of dual-energy carotid CT angiography under dual-low scan.
PubMed 文献 · Neuroradiology · 2025-12-17

前瞻性研究纳入100例接受双能颈动脉CT血管成像的患者,对照组50例采用噪声指数4和ASIR-V 50%重建,实验组50例采用噪声指数11并分别以ASIR-V 50%、DLIR-M和DLIR-H重建。DLIR-H在所有解剖层面图像噪声最低、SNR和CNR最高,主观评分显著更优,各组血管CT值无显著差异。

要点

  • 双低扫描条件下,40 keV虚拟单能图像采用DLIR-H重建可获得最低噪声和最高SNR、CNR。
  • DLIR-H主观图像清晰度、噪声抑制和整体诊断信心评分显著高于ASIR-V 50%和DLIR-M。
  • 各重建方法间血管CT值无统计学显著差异。

核心结论

  • 在双低扫描条件下,DLIR重建可改善双能颈动脉CT血管成像40 keV虚拟单能图像的图像质量,降低图像噪声并提高SNR和CNR,在降低辐射剂量的同时为诊断提供最佳图像质量。

研究设计与样本

  • 前瞻性研究,于2024年12月至2025年1月在本机构纳入100例接受双能颈动脉CT血管成像的患者;对照组50例采用噪声指数4扫描并以ASIR-V 50%重建,实验组50例采用噪声指数11扫描并分别以ASIR-V 50%、DLIR-M和DLIR-H重建,同一50例患者贡献三个实验亚组。

关键影像表现/征象

  • DLIR-H在所有解剖层面(主动脉弓、锁骨下动脉、颈总动脉、颈内动脉)均显示最低图像噪声和最高SNR、CNR(P < 0.05)。
  • DLIR-H主观评分显著更高,图像清晰度、噪声降低和整体诊断信心更优(P < 0.05)。
  • 各组间血管CT值无统计学显著差异。

诊断标准与关键数值

  • 客观图像质量在主动脉弓、锁骨下动脉、颈总动脉和颈内动脉四个解剖层面测量CT值(HU)、图像噪声(SD)、SNR和CNR。
  • 主观图像质量由两名经验丰富的放射科医师采用五分Likert量表独立评估,聚焦噪声、分辨率和整体图像质量。
  • DLIR-H在所有解剖层面图像噪声最低、SNR和CNR最高(P < 0.05)。
  • DLIR-H主观评分显著更高(P < 0.05)。
  • 各组血管CT值无统计学显著差异。

临床应用与阅片要点

  • 双低扫描方案下,40 keV虚拟单能图像采用DLIR-H重建可在降低辐射剂量的同时获得最优图像质量。
  • 阅片时DLIR-H图像噪声更低、SNR和CNR更高,可能有助于提升诊断信心。

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

OBJECTIVES: This study aimed to assess and compare the efficacy of deep learning image reconstruction (DLIR) and adaptive statistical iterative reconstruction V (ASIR-V) in improving the image quality of 40 keV virtual monochromatic images (VMIs) in dual-energy CT carotid angiography (DE-CTA) under a dual-low scan protocol. METHODS: A prospective study was conducted between December 2024 and January 2025, involving a total of 100 patients who underwent DE-CTA at our institution. Patients were assigned into two distinct cohorts: Control Group (n = 50), scanned with a noise index (NI) of 4 and reconstructed using ASIR-V 50%; and Experimental Group (n = 50), scanned with a noise index (NI) of 11. For the Experimental Group, images were reconstructed using three different methods: ASIR-V 50%, DLIR at a medium-strength setting (DLIR-M), and DLIR at a high-strength setting (DLIR-H). Accordingly, the 50 patients in the Experimental Group were further categorized into Experimental Group 1 (EG 1) for ASIR-V 50%, Experimental Group 2 (EG 2) for DLIR-M, and Experimental Group 3 (EG 3) for DLIR-H. Thus, the same 50 patients contributed to all three experimental subgroups, while the Control Group remained independent with its 50 patients. Objective image quality was assessed at four anatomical levels (aortic arch, subclavian artery, common carotid artery, and internal carotid artery), with measurements of CT values (HU), image noise (SD), signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR). Subjective image quality was independently evaluated by two experienced radiologists using a five-point Likert scale, focusing on noise, resolution, and overall image quality. RESULTS: Among all reconstruction methods, DLIR-H yielded the lowest image noise and the highest SNR and CNR at all anatomical levels (P < 0.05). Subjective evaluation scores were significantly higher for DLIR-H, indicating superior image clarity, noise reduction, and overall diagnostic confidence (P < 0.05). No statistically significant differences were observed in vascular CT values among the groups. CONCLUSION: DLIR reconstruction improves the image quality of 40 keV VMIs in DE-CTA under dual-low scanning conditions, with a reduction in image noise and enhanced SNR and CNR, providing optimal image quality for diagnostic purposes at reduced radiation doses.

来源:Neuroradiology,查看原文。版权归原作者所有。

引用

  1. Qiu S, Xu K, Meng Y, et al. Comparison of deep learning reconstruction algorithms to improve image quality of dual-energy carotid CT angiography under dual-low scan. Neuroradiology. 2025. PMID: 41405722. doi:10.1007/s00234-025-03803-x

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。