路涛,张天悦,李缃琦,郭爱文,宋彬,刘思耘.基于T2WI影像组学模型产前预测胎盘植入性病变[J].中国医学影像技术,2021,37(12):1854~1859
基于T2WI影像组学模型产前预测胎盘植入性病变
Radiomics model based on T2WI for prenatal predicting placenta accreta spectrum disorders
投稿时间:2021-02-17  修订日期:2021-07-29
DOI:10.13929/j.issn.1003-3289.2021.12.022
中文关键词:  胎盘疾病  胎盘植入  磁共振成像  影像组学
英文关键词:placenta diseases  placenta accrete  magnetic resonance imaging  radiomics
基金项目:
作者单位E-mail
路涛 四川省医学科学院·
四川省人民医院放射科, 四川 成都 610072 
 
张天悦 四川省医学科学院·
四川省人民医院放射科, 四川 成都 610072 
 
李缃琦 四川省医学科学院·
四川省人民医院放射科, 四川 成都 610072 
 
郭爱文 四川省医学科学院·
四川省人民医院放射科, 四川 成都 610072 
 
宋彬 四川大学华西医院放射科, 四川 成都 610037 songb_radiology@163.com 
刘思耘 GE医疗, 北京 100176  
摘要点击次数: 1174
全文下载次数: 360
中文摘要:
      目的 评估基于MR T2WI影像组学模型产前预测胎盘植入性病变(PAS)的价值。方法 回顾性分析241例孕妇及胎儿MRI,其中116例PAS、125例无PAS。按7:3比例将其分为训练集(n=168)和验证集(n=73),于训练集提取并筛选半傅立叶采集单次激发快速自旋回波(HASTE)及真实稳态进动快速成像(TrueFISP)序列图像的影像组学特征,构建预测PAS的影像组学模型,并以回归分析方法构建临床模型、影像组学模型及临床-影像组学模型。采用校准曲线和受试者工作特征(ROC)曲线分析模型的效能,以决策曲线分析(DCA)评估其临床实用性。结果 对各序列图像分别提取1 130个影像组学特征,经LASSO回归等处理后,各筛选出9个影像组学特征,用于构建预测PAS的HASTE及TrueFISP影像组学模型。ROC曲线显示,临床模型、HASTE影像组学模型及TrueFISP影像组学模型在验证集中诊断PAS的曲线下面积(AUC)分别为0.882、0.968和0.930(P均>0.05);HASTE联合TrueFISP影像组学模型的AUC为0.990,高于临床(Z=-2.36,P=0.02)、HASTE影像组学(Z=-2.48,P=0.02)及TrueFISP影像组学模型(Z=-2.43,P=0.02);临床-HASTE-TrueFISP影像组学模型的AUC为0.995,与HASTE联合TrueFISP影像组学模型差异无统计学意义(Z=-0.85,P=0.40),高于HASTE或TrueFISP影像组学模型(Z=-2.64、-2.47,P均<0.05)。临床模型之外,各模型在验证集数据中的校准度均较好;阈值取0~0.6时,其在验证集的临床净获益均大于临床模型。结论 基于产前HASTE及TrueFISP序列图像的联合影像组学模型有助于准确预测PAS。
英文摘要:
      Objective To observe the value of radiomics model based on T2WI for prenatal predicting placenta accreta spectrum disorders (PAS). Methods Placental MRI data of 241 pregnant women were retrospectively reviewed, including 116 with PAS (PAS group) and 125 without PAS (non-PAS group). All pregnant women were divided into training set (n=168) and verification set (n=73) at the ratio of 7:3. Radiomics features of half-Fourier acquisition single-shot turbo spin-echo (HASTE) and true fast imaging with steady-state precession (TrueFISP) sequences images were extracted and screened in the training set, and the radiomics models for predicting PAS were constructed. Then clinical, radiomics and clinical-radiomics models were conducted with Logistic regression analysis. The performances of the models were analyzed using calibration curves and receiver operating characteristic (ROC) curves, and the decision curve analysis (DCA) was used to evaluate the clinical practicability of the models. Results Totally 1 130 radiomics features were extracted from each sequence, and 9 features were selected for establishing HASTE or TrueFISP radiomics model for predicting PAS with LASSO method. ROC curve showed that the area under the curve (AUC) of the clinical, HASTE radiomics model and TrueFISP radiomics model was 0.882, 0.968 and 0.930 in verification set, respectively (all P>0.05), of HASTE combined with TrueFISP radiomics model was 0.990, higher than that of clinical model (Z=-2.36, P=0.02), HASTE radiomics model (Z=-2.48, P=0.02) and TrueFISP radiomics model (Z=-2.43, P=0.02). AUC of the clinical-HASTE-TrueFISP radiomics model was 0.995, not significant different with that of HASTE combined with TrueFISP radiomics model (Z=-0.85, P=0.40), but higher than that of HASTE radiomics model and TrueFISP radiomics model (Z=-2.64, -2.47, both P<0.05). All models except for the clinical model had good calibration in the validation set, and the net clinical benefits of them were all greater than that of the clinical model in validation set when the threshold ranged from 0 to 0.6. Conclusion The combined radiomics model based on prenatal HASTE and TrueFISP images was helpful to accurate prediction of PAS.
查看全文  查看/发表评论  下载PDF阅读器