首页|人工智能技术预测胎盘功能不全的可行性研究

人工智能技术预测胎盘功能不全的可行性研究

扫码查看
目的:探索人工智能(AI)技术在孕早期预测胎盘功能不全(PI)患者的可行性.方法:选择单胎前壁胎盘的高危孕妇为研究对象.孕早期收集超声影像数据:子宫动脉搏动指数(UTPI)和胎盘血管血流化参数(PVIs).结局发展为子痫前期(PE)和胎儿生长受限(FGR)的患者被定义为PI.孕早期胎盘图像扩增后分为训练集与测试集,从训练集图像中提取胎盘纹理特征创建基于超声图像技术的胎盘功能不全机器学习模型,用测试集图像进行验证,比较超声与AI的诊断试验结果.结果:纳入164例孕妇,胎盘功能正常组和PI组分别为147例和17例,AI测试集对预测PI的敏感度、特异度、阳性预测值、阴性预测值、阳性似然比和阴性似然比分别为73.33%、60.85%、58.93%、74.87%、1.87和0.44,受试者工作特征(ROC)曲线下面积0.67,精确率58.93%,召回率73.33%,F1分数0.65.超声与AI的诊断试验结果比较:AI阳性预测值最高(58.93%),其他诊断试验指标均低于PVIs.结论:基于胎盘超声纹理分析的AI技术的阳性预测值优于超声参数,鉴于PVIs值容易受到许多技术或生理性参数的影响,AI是预测PI崭新有希望的工具.
Feasibility Study of Artificial Intelligence Technology to Predict Placental Insuf-ficiency
Objective:To explore the feasibility of artificial intelligence(AI)technology in predicting placental in-sufficiency in the first trimester.Methods:High risk pregnant women with singleton and anterior wall placenta were selected as the research subjects.Data of ultrasound parameters were collected in the first trimester,including u-terine arterial pulsatility index(UTPI)and placental vascular hemodynamics index(PVIs).Patients that developed into preeclampsia(PE)and fetal growth restriction(FGR)were defined as the placental insufficiency group.The placental images in the first trimester were asymmetrically amplified and divided into a training set and a testing set.The placental texture features were extracted from the training set images to create a placental insufficiency machine learning model based on ultrasound imaging technology.The model was validated using the testing set images,and the diagnostic test results of ultrasound and AI were compared.Results:A total of 164 pregnant women were included,with 147 normal cases and 17 cases with placental insufficiency.The sensitivity,specificity,positive predictive value,negative predictive value,positive likelihood ratio and negative likelihood ratio of the AI test set were 73.33%,60.85%,58.93%,74.87%,1.87 and 0.44,respectively.The area under the receiver operat-ing characteristic(ROC)curve was 0.67,with an accuracy rate of 58.93%,a recall rate of 73.33%,and an F1 score of 0.65.The comparison of diagnostic test results between ultrasound and AI:AI had the highest positive predictive value(58.93%),while other diagnostic test results were lower than PVIs.Conclusions:The positive predictive value of AI technology based on placental ultrasound texture analysis is superior to ultrasound parame-ters.Given that PVIs values are easily affected by many technical or physiological parameters,AI is a promising tool for predicting placental dysfunction.

Placental insufficiencyArtificial intelligenceUterine artery pulsatility indexPlacental volumeVas-cularization flow index

陈结云、陈敏、陈敦金

展开 >

广州医科大学附属第三医院:产前诊断(胎儿医学)科,广东 广州 510150

广州医科大学附属第三医院:人工智能实验室,广东 广州 510150

广州医科大学附属第三医院:广东省产科重大疾病重点实验室广东省妇产疾病临床医学研究中心 粤港澳母胎医学高校联合实验室,广东 广州 510150

胎盘功能不全 人工智能 子宫动脉搏动指数 胎盘体积 胎盘血管血流化参数

2024

实用妇产科杂志
四川省医学会

实用妇产科杂志

CSTPCD北大核心
影响因子:2.564
ISSN:1003-6946
年,卷(期):2024.40(12)