Page 39 - 磁共振成像2024年7期电子刊

特别关注｜｜Special Focus 磁共振成像 2024年7月第15卷第7期 Chin J Magn Reson Imaging, Jul, 2024, Vol. 15, No. 7
PET/MR专题｜｜临床研究
18 F-FDG PET/MR影像组学特征与宫颈癌PD-L1

表达的相关性研究

李旺，李朗俊，刘卓男，孙洪赞 *

作者单位 中国医科大学附属盛京医院放射科，沈阳 110022
* 通信作者 孙洪赞，E-mail: sunhongzan@126.com
中图分类号 R445.2；R737.33 文献标识码 A DOI 10.12015/issn.1674-8034.2024.07.006
18
本文引用格式 李旺, 李朗俊, 刘卓男, 等 . F-FDG PET/MR 影像组学特征与宫颈癌 PD-L1 表达的相关性研究[J]. 磁共振成像,
2024, 15(7): 32-38, 45.
18
[摘要] 目的 探讨 F-氟代脱氧葡萄糖 （fluorodeoxyglucose, FDG） 正电子发射断层成像 （positron emission tomography,
PET）/MR 影像组学特征与宫颈癌程序性死亡受体配体 1（programmed death-ligand 1, PD-L1）表达的相关性。材料与方法 回
18
顾性分析中国医科大学附属盛京医院 2017年 5月至 2023年 7月进行了 F-FDG PET/MR扫描的 26例宫颈癌患者临床和影像资料，
在 PET，T1WI及 T2WI的图像中对原发灶进行感兴趣区（region of interest, ROI）勾画，将每个患者的每个包含 ROI横截面作为
样本，以手术标本取样染色，将样本分为阳性组（n=233，73.97%）、阴性组（n=82，26.03%），并基于一阶统计量（Firstorder）
从图像中提取影像组学特征。采用独立样本 t检验或 Mann-Whitney U 检验比较两组间特征参数的差异，并分析图像特征参数与
PD-L1表达的相关性。将样本按照 7∶3的比例随机划分为训练集和测试集，将差异有统计学意义的组学特征作为逻辑回归的参
数建立 PET 影像组学模型、MR 影像组学模型及 PET/MR 联合模型，通过受试者工作特征（receiver operating characteristic,
ROC） 曲线的曲线下面积 （area under the curve, AUC） 分析评估各模型对 PD-L1 表达诊断的效能。结果 PET 图像参数
10Percentile（P<0.01）、90Percentile（P<0.01）、Energy（P<0.01）、Interquartile Range（P<0.01） 等 15 个一阶统计量特征与
PD-L1 表达具有较强的相关性，T1WI 图像参数 10Percentile（P<0.01）、90Percentile（P<0.05）、Maximum（P<0.05）、Mean
（P<0.01）等 9 个一阶统计量特征与 PD-L1 表达具有较强的相关性，T2WI 图像参数 Entropy（P<0.05）、Skewness（P<0.01）、
Energy（P<0.05）、Interquartile Range（P<0.01）等 9 个一阶统计量特征与 PD-L1 表达具有较强的相关性。在影像组学模型中，
训练集 PET 组学模型、MR 组学模型及 PET/MR 联合模型 AUC 值分别为 0.478 [95% 置信区间（confidence interval, CI）：0.389~
0.565]、0.806（95% CI：0.728~0.874）、0.850（95% CI：0.784~0.909）。测试集 PET 组学模型、MR 组学模型及 PET/MR 联合模
型 AUC 值 分 别 为 0.528 （95% CI： 0.381~0.669）、 0.737 （95% CI： 0.623~0.843）、 0.817 （95% CI： 0.715~0.902）。
18
结论 F-FDG PET/MR影像组学特征与宫颈癌 PD-L1表达差异有较强的相关性，PET/MR影像组学模型在预测宫颈癌 PD-L1表
达上具有更好的效能，能在临床上为宫颈癌患者评价PD-L1的表达以优化个体诊疗方案，改善患者预后。
[关键词] 妇科肿瘤；宫颈癌；程序性死亡受体配体1；影像组学；磁共振成像；正电子发射断层成像

18
Correlation study between F-FDG PET/MR imaging radiomic features and PD-L1 expression
in cervical cancer
LI Wang, LI Langjun, LIU Zhuonan, SUN Hongzan *
Department of Radiology, Shengjing Hospital of China Medical University, Shenyang 110022, China
* Correspondence to SUN H Z, E-mail: sunhongzan@126.com
Received 16 Jan 2024, Accepted 17 Apr 2024; DOI 10.12015/issn.1674-8034.2024.07.006
ACKNOWLEDGMENTS Shenyang Young and Middle-aged Scientist Science and Technology Innovation Talent Program (No. RC210138).
18
Cite this article as LI W, LI L J, LIU Z N, et al. Correlation study between F-FDG PET/MR imaging radiomic features and PD-L1
expression in cervical cancer[J]. Chin J Magn Reson Imaging, 2024, 15(7): 32-38, 45.
Abstract Objective: To explore the correlation between F-fluorodeoxyglucose (FDG) positron emission tomography (PET)/MR
18
radiomic features and the expression of programmed death-ligand 1 (PD-L1) in cervical cancer. Materials and Methods: A retrospective
18
analysis was conducted on 26 cervical cancer patients who underwent F-FDG PET/MR scans at Shengjing Hospital, China Medical
University, from May 2017 to July 2023. Regions of interest (ROIs) were delineated on the images of the primary lesions in PET, T1WI,
and T2WI. Each cross-sectional image containing an ROI for each patient was treated as a sample. Based on sampling and staining of
surgical specimens, the samples were divided into a positive group (n=233, 73.97%) and a negative group (n=82, 26.03%). Radiomic
features were extracted from the images using first-order statistics. Independent sample t-tests or Mann-Whitney U tests were used to
compare the differences in feature parameters between the two groups. The correlation between image feature parameters and PD-L1
expression was analyzed. The samples were randomly divided into training and testing sets in a 7∶ 3 ratio. Radiomic features with
statistically significant differences were used as parameters to establish PET radiomic models, MR radiomic models, and PET/MR
combined models through logistic regression. The diagnostic performance of each model for PD-L1 expression was evaluated using the
area under the curve (AUC) analysis of receiver operating characteristic (ROC) curves. Results: The 15 first-order statistical features of
PET images, including 10Percentile (P<0.01), 90Percentile (P<0.01), Energy (P<0.01), Interquartile Range (P<0.01), and others, exhibit
a strong correlation with PD-L1 expression. Similarly, the T1WI image parameters, such as 10Percentile (P<0.01), 90Percentile (P<
0.05), Maximum (P<0.05), Mean (P<0.01), and nine other first-order statistical features, show a strong correlation with PD-L1
expression. Additionally, the T2WI image parameters, including Entropy (P<0.05), Skewness (P<0.01), Energy (P<0.05), Interquartile Range (P<
18
0.01), and eight other first-order statistical features, demonstrate a strong correlation with PD-L1 expression. Conclusions: F-FDG PET/MR

收稿日期 2024-01-16 接受日期 2024-04-17
基金项目 沈阳市中青年科技创新人才支持计划项目（编号：RC210138）

·32 · https://www.chinesemri.com