Page 72 - 南京医科大学自然版

南京医科大学学报（自然科学版） 第45卷第6期
·810 · Journal of Nanjing Medical University（Natural Sciences） 2025年6月

·临床研究·

深度学习重建算法在胰腺HASTE⁃T2WI序列中的临床应用价值

徐思雨 ，张永杰 ，田 水 ，王建伟 1*
1
1
2
南京医科大学第一附属医院放射科，江苏 南京 210029；南京医科大学基础医学院人体解剖学系，江苏 南京 211166
1 2

［摘 要］ 目的：探讨基于深度学习（deep learning，DL）重建算法的半傅立叶采集单次快速自旋回波（half⁃Fourier acquisition
single⁃shot turbo spin echo，HASTE）T2WI序列在胰腺磁共振成像（magnetic resonance imaging，MRI）中的临床应用价值。方法：
应用3.0T 磁共振对69例患者进行胰腺常规BLADE⁃TSE⁃T2WI和基于DL的HASTE⁃DL⁃T2WI序列扫描。采用Likert Scale量表
5分法对两组图像总体质量、胰腺锐利度、胆管显示清晰度、伪影进行主观评分；测量并比较两组图像同层面的胰腺正常组织、
病灶的对比噪声比（contrast to noise ratio，CNR）和信噪比（signal to noise ratio，SNR），并记录扫描时间。结果：HASTE⁃DL 序列
的图像总体质量、胰腺锐利度以及胆管显示清晰度评分均优于BLADE⁃TSE序列（P < 0.001），两组图像伪影评分差异无统计学
意义（P > 0.05）。HASTE⁃DL 序列图像的正常胰腺组织 SNR、病灶 SNR 和 CNR 均优于 BLADE⁃TSE 序列图像（P < 0.001），
HASTE⁃DL序列的扫描时间较 BLADE⁃TSE 序列缩短了 78%。结论：与 BLADE⁃TSE 序列图像相比，HASTE⁃DL 序列图像总体
质量更好，胰腺锐利度及胆管显示清晰度更高、SNR 和 CNR 更优，并且扫描时间更短，在胰腺 MRI 扫描中有很好的临床应
用价值。
［关键词］ 胰腺；磁共振成像；图像质量；深度学习
［中图分类号］ R445.2 ［文献标志码］ A ［文章编号］ 1007⁃4368（2025）06⁃810⁃07
doi：10.7655/NYDXBNSN241084

The clinical value of deep learning reconstruction algorithm in pancreatic HASTE⁃T2WI
sequence

1
1
2
XU Siyu ，ZHANG Yongjie ，TIAN Shui ，WANG Jianwei 1*
1 Department of Radiology，the First Affiliated Hospital of Nanjing Medical University，Nanjing 210029；Department
2
of Human Anatomy，School of Basic Medicine，Nanjing Medical University，Nanjing 211166，China
［Abstract］ Objective：To evaluate the clinical value of deep learning（DL）⁃ based reconstruction algorithm for half ⁃ Fourier
acquisition single ⁃ shot turbo spin echo（HASTE）T2 ⁃ weighted imaging（T2WI）in pancreatic magnetic resonance imaging（MRI）.
Methods：A total of 69 patients underwent pancreatic conventional BLADE⁃TSE⁃T2WI and based on deep learning HASTE⁃DL⁃T2WI
sequences scanning using 3.0T MR. The overall image quality，pancreatic sharpness，biliary duct clarity，and artifacts were
subjectively scored using a Likert 5⁃point scale. The contrast to noise ratio（CNR）and signal to noise ratio（SNR）of normal pancreatic
tissue and the lesion in both sequences were measured and compared，and the scan time was recorded. Results：The HASTE⁃DL
sequence scored significantly higher in overall image quality，pancreatic sharpness，and bile duct clarity than the BLADE ⁃ TSE
sequence（P < 0.001），with no statistical difference in artifact scores（P > 0.05）. The SNR of normal pancreatic tissue，lesion SNR，and
CNR in HASTE⁃DL images were superior to those of BLADE⁃TSE sequence（P < 0.001）. Additionally，the scanning time of HASTE⁃
DL was reduced by 78% compared to BLADE⁃TSE. Conclusion：Compared to BLADE⁃TSE sequence，HASTE⁃DL provides better
overall image quality，superior pancreatic sharpness and bile duct clarity，higher SNR and CNR，and significantly shorter scan time.
Thus，HASTE⁃DL T2WI demonstrates excellent clinical utility in pancreatic MRI.
［Key words］ pancreas；magnetic resonance imaging；image quality；deep learning
［J Nanjing Med Univ，2025，45（06）：810⁃815，825］

［基金项目］ 江苏省青蓝工程中青年学术带头人基金（KY101R202023）
通信作者（Corresponding author），E⁃mail：wangjianwei@jsph.org.cn（ORCID：0009⁃0002⁃5807⁃1437）
∗