影像科学与光化学2025，Vol.43Issue(6)：84-95,12.DOI:10.7517/issn.1674-0475.2025.06.12
基于多切面融合识别技术的胎儿圆锥动脉干畸形智能化超声诊断研究
Research on Intelligent Ultrasonic Diagnosis of Fetal Conotruncal Defects Based on Multi Sectional Fusion Recognition Technology
摘要
Abstract
Objective:This study aims to develop and validate an intelligent ultrasound diagnostic system based on multi sectional fusion recognition technology for early screening and accurate diagnosis of fetal conotruncal defects(CTD),in order to improve the detection rate and diagnostic accuracy of fetal heart diseases.Methods:A total of 421 singleton pregnant women at 18-32 weeks of gestation who underwent grade III prenatal ultrasound and fetal echocardiography from January 1,2019 to September 30,2024 were retrieved from the ultrasound database of Dongguan Maternal and Child Health Care Hospital.From these,2895 images-including four chamber views(4CH),three vessel and tracheal views(3VT),right ventricular outflow tract(RVOT),left ventricular outflow tract(LVOT),and blood flow Doppler images of normal or abnormal fetuses were used as the training set.The fetal echocardiography section was extracted using the classic YOLOv3 network through a series of convolutional layers to identify eight major clinical screening events,namely 4CH,3VT The grayscale ultrasound timing and color Doppler timing under LVOT and RVOT were analyzed by combining the expert designed diagnostic gold standard,training a set of neural networks,and integrating modeling.Another 250 pregnant women were selected as the validation set:Group A(200 normal pregnancies),and Group B(50 pregnancies with fetal arterial trunk abnormalities).The accuracy,sensitivity,and specificity of doctor+multi section fusion recognition technology(AI+doctor)diagnosis and AI alone diagnosis on different sections of different datasets,and the time consumption of the two diagnostic methods were statistically analyzed.Results:The research results showed that the diagnostic accuracy,sensitivity,and specificity of the 3VT section were the highest.The accuracy,sensitivity,and specificity of different sectional diagnoses in Group B were significantly higher than those in Group A(P<0.05).In addition,there was no significant difference in accuracy,sensitivity,and specificity between AI diagnosis alone and AI+doctor diagnosis(P>0.05).Individual AI diagnosis took 195 seconds,while AI+doctor diagnosis took 1560 seconds.The efficiency of individual AI diagnosis was significantly higher than that of AI+doctor diagnosis(P<0.05).Conclusion:The diagnostic effect of intelligent ultrasound diagnosis based on multi section fusion recognition technology is not significantly different from AI+doctor diagnosis,but the time consumption is significantly reduced,showing significant advantages.关键词
胎儿圆锥动脉干畸形/多切面融合识别技术/智能化超声诊断/深度学习/图像处理Key words
fetal conotruncal defects/multi sectional fusion recognition technology/intelligent ultrasound diagnosis/deep learning/image processing分类
医药卫生引用本文复制引用
徐秋红,郭红梅,邓政喜,任双丽,汪南..基于多切面融合识别技术的胎儿圆锥动脉干畸形智能化超声诊断研究[J].影像科学与光化学,2025,43(6):84-95,12.基金项目
2022年东莞市社会发展科技项目(20221800905722). （20221800905722）
