基于卷积神经网络的肠道寄生虫虫卵识别模型研究  

Research on recognition model of intestinal parasite eggs based on convolutional neural network

作　　者：殷李华 胡建雄 方悦怡[3] 容祖华 黄栩滨 何冠豪 江芷莹 肖建鹏[4] 刘涛[1] 马文军 YIN Lihua;HU Jianxiong;FANG Yueyi;RONG Zuhua;HUANG Xubin;HE Guanhao;JIANG Zhiying;XIAO Jianpeng;LIU Tao;MA Wenjun(Department of Public Health and Preventive Medicine,School of Medicine,Jinan University,Guangzhou 510632,China;School of Information Engineering,Guangzhou City Construction College;Parasitic Disease Prevention and Control Institute,Guangdong Provincial Center for Disease Control and Prevention;Guangdong Provincial Institute of Public Health,Guangdong Provincial Center for Disease Control and Prevention)

机构地区：[1]暨南大学基础医学与公共卫生学院公共卫生与预防医学系,广东广州510632 [2]广州城建职业学院信息工程学院 [3]广东省疾病预防控制中心寄生虫病预防控制所 [4]广东省疾病预防控制中心广东省公共卫生研究院

出　　处：《华南预防医学》2023年第12期1498-1503,共6页South China Journal of Preventive Medicine

基　　金：广东省自然科学基金项目(2021A1515012578)。

摘　　要：目的构建肠道寄生虫卵的粪检显微图像数据集,建立一个深度学习模型,为肠道寄生虫疾病辅助诊断提供技术支撑。方法利用显微镜和数码相机采集12种肠道寄生虫虫卵显微图像,经预处理后对虫卵的类别和位置进行标注,形成粪检显微图像数据集。以掩膜区域卷积神经网络深度学习模型作为框架,对标定框回归、分类、掩膜进行训练,并评估其性能。结果构建的图像数据集共6299张图片,涵盖了10944个虫卵图像。经测试建立的深度学习模型总体识别准确率为90.20%,12种虫卵的准确率为58.65%(曼氏迭宫绦虫卵)~100.00%(蛲虫卵)。结论构建肠道寄生虫卵的显微图像数据集和利用卷积神经网络建立肠道寄生虫卵显微图像的识别模型可为寄生虫相关疾病的辅助诊断提供技术支撑。Objective To construct a fecal microscopy image dataset of intestinal parasite eggs and establish a corresponding deep learning image recognition model,so as to provide technical support for the auxiliary diagnosis of intestinal parasitic diseases.Methods Microscopic images of 12 intestinal parasite eggs were collected using a microscope and a digital camera,pre-processed and labelled with the categories and locations of the eggs to form an image dataset.A masked region convolutional neural network deep learning model was used as a framework to train the bounding box regression,classification,and mask generation,and its performance was evaluated.Results The fecal microscopy image dataset was constructed with a total of 6299 images,including 10944 egg images.After testing,the deep learning model achieved an overall recognition accuracy of 90.20%.The recognition accuracy for the 12 types of eggs ranged from 58.65%(Spirometra mansoni egg)to 100.00%(Pinworm egg).Conclusion Constructing a microscopy image dataset of intestinal parasite eggs and establishing an image recognition model using convolutional neural networks provide technical support for the auxiliary diagnosis of parasitic diseases.

关 键 词：图像数据集 肠道寄生虫 肠道寄生虫卵 深度学习 

分 类 号：R183.9[医药卫生—流行病学] R195.1[医药卫生—公共卫生与预防医学]