人工智能自动识别胎儿颜面部超声标准切面的研究

Automatic recognition of fetal facial ultrasound standard plane using artificial intelligence

目的:探讨人工智能(AI)自动识别与分类胎儿颜面部超声标准切面(FFUSP)的价值。方法:以妊娠20~24周FFUSP图像为研究对象,含标准集1906张和实验集4532张。标准集分为训练集和测试集用于训练和测试AI模型识别分类鼻唇冠状切面、正中矢状面、经双眼球横切面及非标准切面;以产科超声专家分类为标准,比较分析AI、初级医生组、中级医生组对实验集FFUSP图像识别分类能力差异。结果:AI对测试集各切面的分类准确率均达97%以上,中级医生对实验集FFUSP各切面识别能力皆优于初级医生(P<0.05)。AI对FFUSP各切面总体识别效能优于初级医生和中级医生(P<0.05),与专家分类一致性强(P<0.05);AI分类效率显著优于医生人工分类(P<0.05)。结论:AI对FFUSP识别分类具有较高准确性,可作为胎儿超声规范化培训和图像质量控制的辅助方法。

Objective To explore the value of artificial intelligence(AI)for automatically identifying and classifying fetal facial ultrasound standard plane(FFUSP).Methods The FFUSP at 20-24 weeks gestation were taken as the research object,including 1906 images in standard set and 4532 images in experimental set.The images in standard set were further divided into training set and test set for training and testing the ability of AI in recognizing and classifying nasolabial coronal plane,median sagittal plane,ocular axial plane and non-standard plane,respectively.Taking the classification by obstetric ultrasound experts as the standard,the differences among AI,junior doctors and intermediate doctors in the recognition and classification of FFUSP in experimental set were compared and analyzed.Results The classification accuracy of AI on each kind of planes in test set was higher than 97%.Intermediate doctors surpassed junior doctors in the recognition of FFUSP in experimental set(P<0.05).AI was superior to junior doctors and intermediate doctors in the total recognition efficiency of FFUSP(P<0.05),and had a strong consistency with the classification results obtained by experts(P<0.05).The classification efficiency of AI was significantly better than the artificial classification by doctors(P<0.05).Conclusion AI which has a high accuracy in FFUSP identification and classification can be used as an assistant method for fetal ultrasonic standardized training and image quality control.