1. a. Department of Clinical Medical Engineering; b. Department of Obstetrics and Gynecology, Nanjing Drum Tower Hospital, Nanjing
Jiangsu 210008, China; 2. College of Bioscience and Medical Engineering, Southeast University, Nanjing Jiangsu 210096, China
Abstract：Objective To recognize 31 standard sections of fetal ultrasound screening images in mid-pregnancy automatically by
using deep learning method and depth convolution neural network model. Methods From 20 to 24 weeks of gestation, a total of
76260 fetal ultrasound screening images (including 31 sections) were collected. These images were divided into training set (68386)
and test set (7874). On the Vgg16 network model, fine tuning was carried out, and data sets were loaded for training. The trained
model was validated in the test set. Results The recognition accuracy of the model for fetal ultrasound screening section was 94.8%.
Conclusion This method can accurately identify each section of fetal ultrasound screening image, and lay a solid foundation for the
solution of automatic quality control of fetal ultrasound image.
陈健1a,2，赵海桐2
，杨玉志1a，徐志扬1a，茹彤1b. 基于深度学习的中孕期胎儿超声筛查切面自动识别[J]. 中国医疗设备, 2020, 35(5): 105-108.
CHEN Jian1a,2, ZHAO Haitong2
, YANG Yuzhi1a, XU Zhiyang1a, RU Tong1b. Automatic Recognition of Fetal Ultrasound Screening Section in Middle Pregnancy Based on
Deep Learning. China Medical Devices, 2020, 35(5): 105-108.
