超声图像中颈动脉血管自动识别算法研究  

Object Detection of the Carotid Artery in the Ultrasound Image

作　　者：刘琼 戴柯 周然 黄文健 丁明跃[1,2] Liu Qiong;Dai Ke;Zhou Ran;Huang Wenjian;Mingyue Ding(Department of Biomedical Engineering,School of Life Science and Technology,Huazhong University of Science and Technology,Wuhan,Hubei,430074;Ministry of Education Key Laboratory of Molecular Biophysics,Huazhong University of Science and Technology,Wuhan,Hubei,430074)

机构地区：[1]华中科技大学生命科学与技术学院生物医学工程系,湖北武汉430074 [2]华中科技大学分子生物物理教育部重点实验室,湖北武汉430074

出　　处：《生命科学仪器》2019年第6期109-112,108,共5页Life Science Instruments

基　　金：国家自然科学项目(编号:81571754);国家级大学生创新创业计划资助项目(编号:201910487065)。

摘　　要：在多数情况下,基于深度学习进行医学图像分割问题非常依赖于人工标记感兴趣区域(ROI)。针对这个问题,本文提出引入在自然图像目标检测问题上有优秀表现的Faster R-CNN算法进行超声图像中血管ROI的自动检测,平均交并比(IoU)为90.12%。Faster R-CNN算法应用在医学图像目标检测问题上具有训练周期短,消耗内存小,预测速度快的优点。将Faster R-CNN与U-Net相结合能够实现超声图像中血管的完全自动分割,对于开发计算机辅助诊疗系统有临床价值。In most cases,medical image segmentation based on deep learning relies on a manual region of interest(ROI).To solve this problem,a Faster R-CNN algorithm with excellent performance in natural image object detection was introduced for automatic detection of vascular ROI in ultrasound images,with an average intersection ratio(IoU)of 90.12%.The application of the Faster R-CNN algorithm in medical image object detection has the advantages of short training time,small memory consumption and fast prediction speed.The combination of Faster R-CNN and U-Net can realize the complete automatic segmentation of blood vessels in ultrasound images,which is of great clinical value for the development of computer-aided diagnosis.

关 键 词：超声 图像处理 目标检测 计算机辅助诊断 

分 类 号：R445[医药卫生—影像医学与核医学]