椭圆
胎头
人工智能
分割
计算机视觉
计算机科学
边界(拓扑)
周长
背景(考古学)
主管(地质)
成像体模
像素
模式识别(心理学)
数学
几何学
胎儿
光学
物理
地质学
数学分析
地貌学
古生物学
生物
怀孕
遗传学
作者
Jinting Wang,Zhiwen Fang,Sheng Yao,Feng Yang
标识
DOI:10.1016/j.bspc.2022.104535
摘要
Precise fetal head circumference measurement by ultrasound imaging is of great significance for prenatal examination. However, missing or blurring boundaries caused by artifacts and noises challenge measurement accuracy. The inconsistency between the segmentation pseudo-label and the ellipse contours also generates measurement errors. To improve the measurement performances of fetal head circumference, in this study, we propose an ellipse-guided multi-task network that measures the fetal head circumference according to detected ellipse boundary pixels. In the proposed network, an region segmentation branch is designed to learn region features of the fetal head, and a feature fusion module is applied to combine region features with boundary features, which contribute to exploring more context information about the fetal head and locating boundary pixels in boundary missing or blurring regions. A loss function is also designed in the network to ensure the boundary estimation in an ellipse shape. Experiments are conducted on both the public fetal head circumference measurement dataset HC-18 and a self-built ultrasonic phantom dataset. The experimental results show that the proposed method achieves DSC 97.97%, HD 1.22 mm, and ADF 1.85 mm, which demonstrates that the proposed method achieves excellent performance to compete with other state-of-the-art methods in fetal head circumference measurement.
科研通智能强力驱动
Strongly Powered by AbleSci AI