深度学习加速单次屏气腹部HASTE序列的诊断信心与可行性
本研究在66例1.5T肝脏MRI患者中比较单次屏气深度学习重建HASTE(HASTEDL,采集时间0:16分钟)与常规呼吸触发BLADE(4:00分钟)及多次屏气HASTE(1:30分钟)的T2加权成像。HASTEDL整体图像质量良好且优于多次屏气HASTE但低于BLADE,噪声、锐利度及伪影与BLADE相当且优于多次屏气HASTE,病灶检出和表征与BLADE相当且优于多次屏气HASTE。结论认为单次屏气HASTEDL可行,图像质量与诊断信心与BLADE相当,可能显著节省腹部成像时间。
要点
- 单次屏气HASTEDL采集时间仅0:16分钟,远短于BLADE的4:00分钟和多次屏气HASTE的1:30分钟。
- HASTEDL整体图像质量良好(中位3分,四分位距3-4),优于多次屏气HASTE但低于BLADE。
- HASTEDL病灶检出与表征与BLADE相当,且显著优于多次屏气HASTE。
关键影像表现/征象
- HASTEDL整体图像质量中位评分为3分(四分位距3-4),显著优于多次屏气HASTE(P<0.001),低于BLADE(P=0.001)。
- HASTEDL的噪声、锐利度和伪影与BLADE相当(P≤0.176),且显著优于多次屏气HASTE(P<0.001)。
- HASTEDL病灶检出和病灶表征与BLADE相当(P≤0.523),且显著优于多次屏气HASTE(P<0.001)。
- 三种序列在检出病灶数量和最大病灶测量直径上无显著差异(P≤0.912)。
诊断标准与关键数值
- 两名阅片者均将HASTEDL的诊断信心评为优秀,显著优于多次屏气HASTE(P<0.001),低于BLADE(P=0.044)。
- 图像质量、噪声、锐利度、伪影、诊断信心、病灶检出及病灶表征均采用1-4分Likert量表评分,4分为最佳。
临床应用与阅片要点
- 单次屏气HASTEDL可在一次屏气(0:16分钟)内完成腹部T2加权成像,图像质量良好且诊断信心优秀。
- HASTEDL病灶检出和表征与BLADE相当,阅片时可将其作为快速T2加权序列使用。
- HASTEDL整体图像质量低于BLADE(P=0.001),诊断信心亦低于BLADE(P=0.044),阅片时需注意这一差异。
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
OBJECTIVE: The aim of this study was to evaluate the feasibility of a single breath-hold fast half-Fourier single-shot turbo spin echo (HASTE) sequence using a deep learning reconstruction (HASTEDL) for T2-weighted magnetic resonance imaging of the abdomen as compared with 2 standard T2-weighted imaging sequences (HASTE and BLADE).
MATERIALS AND METHODS: Sixty-six patients who underwent 1.5-T liver magnetic resonance imaging were included in this monocentric, retrospective study. The following T2-weighted sequences in axial orientation and using spectral fat suppression were compared: a conventional respiratory-triggered BLADE sequence (time of acquisition [TA] = 4:00 minutes), a conventional multiple breath-hold HASTE sequence (HASTES) (TA = 1:30 minutes), as well as a single breath-hold HASTE with deep learning reconstruction (HASTEDL) (TA = 0:16 minutes). Two radiologists assessed the 3 sequences regarding overall image quality, noise, sharpness, diagnostic confidence, and lesion detectability as well as lesion characterization using a Likert scale ranging from 1 to 4 with 4 being the best. Comparative analyses were conducted to assess the differences between the 3 sequences.
RESULTS: HASTEDL was successfully acquired in all patients. Overall image quality for HASTEDL was rated as good (median, 3; interquartile range, 3-4) and was significantly superior to HASTEs (P < 0.001) and inferior to BLADE (P = 0.001). Noise, sharpness, and artifacts for HASTEDL reached similar levels to BLADE (P ≤ 0.176) and were significantly superior to HASTEs (P < 0.001). Diagnostic confidence for HASTEDL was rated excellent by both readers and significantly superior to HASTEs (P < 0.001) and inferior to BLADE (P = 0.044). Lesion detectability and lesion characterization for HASTEDL reached similar levels to those of BLADE (P ≤ 0.523) and were significantly superior to HASTEs (P < 0.001). Concerning the number of detected lesions and the measured diameter of the largest lesion, no significant differences were found comparing BLADE, HASTES, and HASTEDL (P ≤ 0.912).
CONCLUSIONS: The single breath-hold HASTEDL is feasible and yields comparable image quality and diagnostic confidence to standard T2-weighted TSE BLADE and may therefore allow for a remarkable time saving in abdominal imaging.
引用
- Herrmann J, Gassenmaier S, Nickel D, et al. Diagnostic Confidence and Feasibility of a Deep Learning Accelerated HASTE Sequence of the Abdomen in a Single Breath-Hold. Invest Radiol. 2021. PMID: 33208596. doi:10.1097/RLI.0000000000000743
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。