摘要
目的基于治疗前多参数磁共振成像(multi-parametric magnetic resonance imaging,mpMRI)影像组学特征,结合临床变量构建模型预测宫颈癌(cervical cancer,CC)脉管浸润(lymphovascular space invasion,LVSI)和预后。材料与方法回顾分析125例CC患者病例,采集小视野高分辨率T2加权成像、表观扩散系数图、轴位T2加权成像(T2-weighted imaging,T2WI)压脂序列和矢状位T2WI、轴位和矢状位对比增强T1加权成像。勾画肿瘤区域后提取107个特征,通过最小绝对值压缩与选择算法等降维以建立影像组学分数(radiomics score,Rad-score),整合14个临床指标构建逐步逻辑回归模型,并重复20次3折交叉验证。根据预测的LVSI及随访结果进行分组及相应无进展生存期(progression-free survival,PFS)的生存曲线划分,观察模型在PFS分组的差异。结果形态学和异质性相关的影像组学特征是预测LVSI的主要因素。回归分析确定3个危险因素,Rad-score比鳞状细胞癌抗原和血红蛋白更重要[优势比(odds ratio,OR):2.626、1.061、0.982]。训练集的受试者工作特征曲线下面积为0.823。PFS在模型预测的LVSI组间明显不同(平均PFS:64.8、58.3个月)。结论mpMRI影像组学特征结合临床变量能预测CC患者的LVSI和临床结局,可能在新辅助和手术环境中显示出改善患者风险分层的效用。影像组学特征能够预测预后可能与其反映肿瘤组织的LVSI有潜在关联。
Objective:The surgical outcomes for patients with cervical cancer(CC)are impaired by lymphovascular space invasion(LVSI).We analyzed the predictive efficacy of radiomic features extracted from pretreatment multi-parameter magnetic resonance imaging(mpMRI)to predict LVSI and clinical outcomes in CC patients due to the lack of a reliable indicator to predict LVSI before surgery.Materials and Methods:A retrospective analysis of 125 individuals with CC was performed.We carried out a radiomic-based characterization on the pretreatment mpMRI to develop and validate a noninvasive imaging biomarker capable of distinguishing between LVSI+and LVSI-.The small field of view high-resolution T2-weighted imaging(T2WI),apparent diffusion coefficient(ADC),T2WI,and contrast-enhanced T1-weighted were included in the image modalities.The volume of interest of six different sequence images contained 107 extracted features in total.These features were then chosen using univariate analysis,LASSO,and stepwise logistic regression analysis.A Rad-score and 14 clinical factors were integrated into the combined(COMB)model,a stepwise logistic regression-based prediction model.Twenty times 3-fold cross-validation was repeated.The progression-free survival(PFS)survival curve was divided based on the follow-up results and the predicted LVSI status,and a difference in the model for the PFS grouping was observed.Results:Radiomics related to intratumoral heterogeneity served as the primary indicator for LVSI prediction.The corresponding Rad-score varied considerably depending on the LVSI status(P<0.001).Multivariate logistics identified 3 LVSI risk variables.The Rad-score was more important than squamous cell carcinoma antigen and hemoglobin[odds ratio(OR):2.626,1.061,0.982].The radiomic model has an area under the curve(AUC)in the training cohort of 0.823.The COMB model predicted a substantial difference in PFS between the LVSI+and LVSI-groups(median PFS:64.8 vs.58.3months).Conclusions:The LVSI status and clinical outcome of CC patients could be predicted using radiomics features in combination with mpMRI radiomics and clinical variates.It may show utility for improving patient stratification strategies in neoadjuvant and surgical settings.The potential of radiomic features to predict tumor prognosis may be connected to their capacity to reflect the histology of LVSI.
作者
崔雅琼
黄刚
王莉莉
任嘉梁
赵莲萍
周星
马颖
CUI Yaqiong;HUANG Gang;WANG Lili;REN Jialiang;ZHAO Lianping;ZHOU Xing;MA Ying(Department of Radiology,Gansu Provincial Hospital,Lanzhou 730000,China;GE Healthcare China,Shanghai 200203,China)
出处
《磁共振成像》
CAS
CSCD
北大核心
2023年第2期73-82,共10页
Chinese Journal of Magnetic Resonance Imaging
基金
甘肃省卫生健康行业科研项目(编号:GSWSKY2020-15)
甘肃省人民医院院内科研基金(编号:20GSSY1-18)。
关键词
子宫颈肿瘤
淋巴血管间隙浸润
磁共振成像
影像组学
预后
uterine cervical neoplasms
lymphovascular space invasion
magnetic resonance imaging
radiomics
prognosis
