超声图像中胎儿头围的自动测量
李 璟1,倪 东1,李胜利2,韩 笑1,尹晓浪1,汪天富1,陈思平1

1)医学超声关键技术国家地方联合工程实验室, 广东省生物医学信息检测与超声成像重点实验室, 深圳大学医学院,深圳 518060; 2)深圳市妇幼保健院超声科,深圳 518028

图像处理; 超声图像处理; 相位对称性; 边缘检测; 椭圆拟合; 随机森林算法; 胎儿头围测量

The automatic ultrasound measurement of fetal head circumference
Li Jing1, Ni Dong1, Li Shengli2, Han Xiao1, Yin Xiaolang1,

Li Jing1, Ni Dong1, Li Shengli2, Han Xiao1, Yin Xiaolang1, Wang Tianfu1, and Chen Siping11)National-Regional Key Technology Engineering Laboratory for Medical Ultrasound, Guangdong Key Laboratory for Biomedical Measurements and Ultrasound Imaging, College of Medicine, Shenzhen University, Shenzhen 518060, P.R.China2)Department of Ultrasound, Shenzhen Maternal and Child Healthcare Hospital, Shenzhen 518028, P.R.China

image processing; ultrasound image processing; phase symmetry(PS); edge detection; ellipse fitting; random forests algorithm; measurement of the fetal head circumference

DOI: 10.3724/SP.J.1249.2014.05455

备注

提出一种超声图像中胎儿头围自动测量的新方法.利用机器学习的随机森林(random forests,RFs)算法自动检测感兴趣区域(region of interest,ROI),通过图像局部相位对称(phase symmetry,PS)检测头围边缘,使用非迭代椭圆拟合算法拟合出头围椭圆.与医生手动拟合测量的结果对比,145个头像的平均相对偏差为-3.86 mm,表明该方法可以鲁棒的自动检测胎儿头围.

A novel learning based method is proposed to automatically measure the fetal head circumference in ultrasound image. Firstly, the random forests classifier is utilized to detect the region of interest(ROI). Then, the fetal head contour in the ROI is segmented using the local phase symmetry-based method. Finally, the non-iterative ellipse fitting method is employed to determine an ellipse on the segmented head contour. Experiments on 145 images show that the mean relative deviation between automatic and manual measurements is -3.86 mm, which demonstrates the feasibility of the proposed method.

·