AI定量超声纹理分析联合子宫动脉PI值对妊娠期糖尿病患者胎儿预后的评估  

Evaluation of Fetal Prognosis in Patients with Gestational Diabetes Mellitus by Al Quantitative Ultrasound Texture Analysis Combined with Uterine Artery PI Value

作　　者：陈菲[1] 聂芳[1] 陈斌娟[1] 何荣霞[2] Chen Fei;Nie Fang;Chen Binjuan;He Rongxia(Ultrasound Medical Center,The Second Hospital of Lanzhou University,Lanzhou 730030,China;Department of Obstetrics,The Second Hospital of Lanzhou University,Lanzhou 730030,China)

机构地区：[1]兰州大学第二医院超声医学中心,兰州市730030 [2]兰州大学第二医院产科,兰州市730030

出　　处：《中国超声医学杂志》2024年第8期909-913,共5页Chinese Journal of Ultrasound in Medicine

摘　　要：目的 基于人工智能(AI)定量超声纹理分析的胎儿肺成熟度(FLM)评估联合母体子宫动脉搏动指数(PI)对妊娠期糖尿病(GDM)患者胎儿预后进行评估。方法 收集符合纳入、排除标准的GDM患者46例,采集以胎儿四腔心切面为基础的完整胎儿胸部声像图,上传进行AI定量超声图像纹理分析以评估FLM,测试结果得出新生儿患呼吸系统疾病如新生儿呼吸窘迫综合征(NRDS)高或低风险,同时测得孕妇子宫动脉PI值。受检孕妇在48 h内择期分娩后,以阿普加评分及脐动脉血气分析为标准评估新生儿是否存在NRDS。运用受试者工作特征(ROC)曲线评价子宫动脉PI值预测NRDS的诊断效能并获得最佳诊断阈值。统计分析AI定量超声纹理分析评估FLM联合子宫动脉PI值在预测NRDS的价值。结果 本研究中GDM孕妇分娩新生儿患NRDS的发生率为43.48%,AI定量超声纹理分析FLM预测其患NRDS的灵敏度为90.00%,特异度为76.92%,准确度为82.60%,阳性似然比为3.90,阴性似然比为0.13,阳性预测值为75.00%,阴性预测值为90.00%,约登指数为0.67。20例新生儿患NRDS组GDM孕妇子宫动脉PI均值高于26例新生儿正常组,差异有统计学意义(t=3.46,P<0.01)。绘制PI值的ROC曲线,得到PI预测新生儿NRDS的最佳诊断阈值为1.28。以AI定量超声纹理分析结果为NRDS高风险联合孕妇子宫动脉PI值≥1.28作为预测NRDS的诊断阈值,其灵敏度为95%,特异度为92.3%,准确度为93.5%。采用χ^(2)检验比较单独AI评估、单独子宫动脉PI值预测以及二者联合的诊断效能,三者间差异均有统计学意义(χ^(2)=7.56,df=2,P<0.05)。二者联合的诊断效能高于单独AI评估、单独子宫动脉PI值预测。结论 基于AI的定量超声图像纹理分析评估GDM患者FLM并联合子宫动脉PI值能提高新生儿NRDS的预测能力,为产科干预与分娩时机的评估提供一定的科学依据。Objective To evaluate the clinical value of artificial intelligence(AI)quantitative ultrasound texture analysis evaluating fetal lung maturity(FLM)combined with uterine artery pulsatility index(PI)value in fetal prognosis prediction in patients with gestational diabetes mellitus(GDM).Methods 46 GDM patients who met the inclusion and exclusion criteria were collected,and the complete fetal chest sonograms based on the four chamber view of the fetus were collected.The quantitative ultrasound image texture analysis of AI was uploaded for FLM evaluation.Getting the test results showed that newborns suffering from respiratory disease such as neonatal respiratory distress syndrome(NRDS)were high or low risk.At the same time,the PI of maternal uterine arteries was measured.After selective delivery within 48 hours,the NRDS was confirmed by Apgar score and blood gas analysis of umbilical artery.Receiver operating characteristic(ROC)curve was used to evaluate the diagnostic efficacy of PI value of uterine artery and to obtain the best diagnostic threshold.Statistical analysis were performed to evaluate the diagnostic value of AI quantitative ultrasound texture analysis evaluating FLM combined with uterine artery PI value in predicting NRDS.Results In this study,the incidence of NRDS in newborn of GDM pregnant women was 43.48%.The sensitivity,specificity,accuracy,positive likelihood ratio,negative likelihood ratio,positive predictive value,negative pre-dictive value and Youden index of AI quantitative ultrasound texture analysis of FLM in predicting NRDS were 90.00%,76.92%,82.60%,3.90,0.13,75.00%,90.00%,and 0.67,respectively.The mean uterine artery PI of 20 GDM pregnant women with NRDS in newborns was higher than that of 26 GDM pregnant women with normal newborns,the difference was statistically significant(t=3.46,P<0.01).The ROC curve of PI value was drawn,and the best diagnostic threshold of NRDS prediction in newborn was 1.28.High risk of NRDS in AI quantitative ultrasound texture analysis combined with PI value≥1.28 i

关 键 词：人工智能 超声诊断 子宫动脉PI值 妊娠期糖尿病 胎肺成熟度 

分 类 号：R714.5[医药卫生—妇产科学] R445.1[医药卫生—临床医学]