摘要
超声心动图像是诊断心脏疾病、分析心脏功能的重要手段,其中左心室大小、形态是判断心脏是否正常的重要参数,而对超声心动图像中左心室进行有效分割是获取左心室大小、形态等参数的关键。在传统U-Net网络基础上引入密集链接,构建一种对左心室超声心动图进行精确分割的深度学习模型。实验结果表明,密集链接的引入可以有效提高分割精度,该模型最终Dice系数为91.76%±1.78%,而传统的U-Net网络Dice系数为83.52%。和全连接网络等方法比较,该方法具有更高的精度。
Echocardiographic image is an important mean to diagnose heart disease and analyze cardiac function.The size and shape of left ventricle are important parameters to judge whether the heart is normal.Effective segmentation of left ventricle in echocardiographic image is the key to obtain the parameters such as left ventricular size and morphology.In this paper,based on the traditional U-Net network,Dense links are introduced to construct a deep learning model for accurate segmentation of left ventricular echocardiography.The experimental results show that the introduction of Dense links can effectively improve the segmentation accuracy.The final Dice coefficient of this model is 91.76%±1.78%,while the Dice coefficient of traditional U-Net network is 83.52%.Compared with the results of Fully connected network and other methods,the proposed method has higher accuracy.
作者
葛帅
严加勇
谢利剑
姜逊渭
GE Shuai;YAN Jia-yong;XIE Li-jian;JIANG Xun-wei(School of Medical Instrument and Food Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China;School of Medical Instrument,Shanghai University of Medicine&Health Science,Shanghai 201318,China;Children's Hospital Affiliated to Shanghai Jiaotong University,Shanghai 200062,China)
出处
《软件导刊》
2021年第2期206-209,共4页
Software Guide
基金
上海市科委西医引导类项目(18411965800)
上海交大医工交叉重点项目(ZH2018ZDA26)。
关键词
左心室分割
超声图像
深度学习
密集连接
U-Net
left ventricular segmentation
ultrasound image
deep learning
dense connection
U-Net
作者简介
葛帅(1996-),男,上海理工大学医疗器械与食品学院硕士研究生,研究方向为生物医学仪器及医学信息技术;通讯作者:谢利剑。
