革兰氏染色细菌显微图像深度学习分类与计数-《中国医学物理学杂志》

文章信息/Info

Title:: Classification and counting of Gram-stained bacteria by deeply learning in micro-image

作者:: 董宇波1; 王蕊1; 赵慧娟2; 张书景3; 1.烟台大学光电信息科学技术学院，山东烟台 264005； 2.滨州医学院公共卫生与管理学院，山东烟台 264005； 3.河北师范大学职业技术学院，河北石家庄 050000

Author(s):: DONG Yubo1; WANG Rui1; ZHAO Huijuan2; ZHANG Shujing3; 1. School of Opto-Electronic Information Science and Technology, Yantai University, Yantai 264005, China 2. School of Public Health and Management, Binzhou Medical College, Yantai 264005, China 3. College of Career Technology of Hebei Normal University, Shijiazhuang 050000, China

Keywords:: Keywords: Gram-stained bacteria classification & counting U-Net ResNet deep learning

摘要:: 菌种和数量是研究菌群失调和疾病预测的重要参数，然而细菌分类和计数工作主要由人工完成，过程繁琐，极易出错，并且耗时费力。本研究提出一种基于图像深度学习的方法对显微图像中的革兰氏阳性杆菌、革兰氏阴性杆菌、革兰氏阳性球菌和革兰氏阴性球菌进行分类。整个算法过程包括分割和分类识别两部分，首先采用U-Net“渐进分割法”对细菌部分和背景部分进行分割；然后将分割后的细菌分别投入ResNet50模型和VGG19模型进行识别和计数。将经过再训练ResNet50模型和VGG19模型的计数结果与人工分类计数标准的结果进行比较，实验结果表明ResNet50模型可以达到人工分类和计数的准确率。

Abstract:: Abstract: Breeds and quantity of bacteria are important parameters for research of dysbacteriosis as well as disease prediction. However, the classification and counting of bacteria was a cumbersome task mainly done by humans, and the process is error-prone, time-consuming and laborious. In this paper, a method based on image deep learning was proposed to classify the four types of bacteria including Gram-positive bacilli, Gram-negative bacilli, Gram-positive cocci and Gram-negative cocci in the microscopic images. The method consists of two major procedures: one is segmentation and the other is classification and identification. First, U-Net "progressive segmentation" was used to segment the bacteria part and the background part. Second, the segmented bacteria were fed into ResNet50 model and VGG19 model for recognition and counting. Finally, the results from retrained ResNet50 model and retrained VGG19 model were compared with the manual classification counting standard, and the results from retrained ResNet50 model were shown to reach the accuracy of manual counting and classification.