肝囊型包虫病超声图影像区域分割算法研究  

Research on Region Segmentation Algorithm of Hepatic Cystic Echinococcosis Ultrasonoscopy

作　　者：王正业 热娜古丽·艾合麦提尼亚孜 王晓荣[2] 米吾尔依提·海拉提 严传波[3] WANG Zhengye;AIHEMAITINIYAZI Renaguli;WANG Xiaorong;HAILATI Miwueryiti;YAN Chuanbo(School of Public Health,Xinjiang Medical University,Urumqi Xinjiang 830011,China;Department of Ultrasonic Diagnosis,The First Affiliated Hospital of Xinjiang Medical University,Urumqi Xinjiang 830011,China;College of Medical Engineering and Technology,Xinjiang Medical University,Urumqi Xinjiang 830011,China)

机构地区：[1]新疆医科大学公共卫生学院,新疆乌鲁木齐830011 [2]新疆医科大学第一附属医院超声诊断科,新疆乌鲁木齐830011 [3]新疆医科大学医学工程技术学院,新疆乌鲁木齐830011

出　　处：《中国医疗设备》2022年第10期18-23,28,共7页China Medical Devices

基　　金：国家自然科学基金(地区项目)(81560294);省部共建中亚高发病成因与防治国家重点实验室开放课题(SKLHIDCA-2020-YG2)。

摘　　要：目的测试Ostu阈值分割、马尔可夫随机场分割和基于深度学习的Poly-YOLO网络模型分割3种方法在肝囊型包虫病超声图像影像区域的分割性能。方法分别使用单尺度图像增强Ostu阈值分割、马尔可夫随机场分割和基于深度学习方法的Poly-YOLO分割网络对肝囊型包虫超声图像中的扇形影像区域进行分割,以去除图像中的干扰信息,并采用Dice相似系数(Dice Similarity Coefficient,DSC)、重叠度(Intersection of Union,IOU)、真阳性率(True Positive Rate,TPR)、豪斯多夫距离(Hausdorff Distance,HD)评价上述3种算法的分割效能。结果Poly-YOLO算法对肝囊型包虫病超声图像具有较好的分割结果,在有效去除非影像区域信息的同时,DSC可达0.80,TPR为0.88,IOU为0.71,HD为2.11。结论相较于基于SSR的Ostu阈值分割方法、马尔可夫随机场图像分割算法,基于深度学习的Poly-YOLO网络能较好地分割出肝囊型包虫病超声图像扇形影像区域,去除图像中的非影像信息,为后续病灶自动分类研究奠定了一定的理论基础。Objective To test the performance of Ostu threshold segmentation method,Markov random field segmentation method and Poly-YOLO network model segmentation method based on deep learning in ultrasound image of hepatic cystic echinococcosis.Methods Three segmentation methods under single scale image enhancement were respectively used to segment the fan-shaped image area in the ultrasound image of cystic liver hydatid to remove the interference information in the image.Dice similarity coefficient(DSC),intersection of union(IOU),true positive rate(TPR)and Hausdorff distance(HD)were used to evaluate the efficiency of the above three algorithms.Results Poly-YOLO algorithm had good segmentation results for ultrasound images of hepatic cystic echinococcosis.While effectively removing non-image area,the DSC was 0.80,TPR was 0.88,IOU was 0.71,and HD was 2.11.Conclusion Compared with the Ostu threshold segmentation method based on SSR and the Markov random field image segmentation algorithm,the Poly-YOLO network based on deep learning can better segment the fan-shaped image area of the ultrasound image of hepatic cystic echinococcosis,remove the non-image information in the image,and lay a certain theoretical foundation for the follow-up study of automatic classification of lesions.

关 键 词：肝囊型包虫病 超声图像 Ostu阈值分割 Poly-YOLO 马尔可夫随机场 

分 类 号：R445.1[医药卫生—影像医学与核医学] TP751.1[医药卫生—诊断学]