骨髓红粒细胞自动识别的深度学习模型  

Deep Learning Model for Automatic Recognition of Erythroid Cells and Granulocyte Cells in Bone Marrow

作　　者：吴汾奇 吕丽丽[2] 吕迪 冯辰彬 施恬 王维 崔红花[2] 周柚[1,3] WU Fenqi;Lü Lili;Lü Di;FENG Chenbin;SHI Tian;WANG Wei;CUI Honghua;ZHOU You(College of Computer Science and Technology,Jilin University,Changchun 130012,China;Second Hospital,Jilin University,Changchun 130012,China;Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education,Jilin University,Changchun 130012,China)

机构地区：[1]吉林大学计算机科学与技术学院,长春130012 [2]吉林大学第二医院,长春130012 [3]吉林大学符号计算与知识工程教育部重点实验室,长春130012

出　　处：《吉林大学学报（信息科学版）》2020年第6期729-736,共8页Journal of Jilin University（Information Science Edition）

基　　金：国家自然科学基金资助项目(61772227,61972174);吉林省科技发展计划重点研发基金资助项目(20180201045GX);吉林大学大学生创新训练计划基金资助项目(201910183289)。

摘　　要：为了实现骨髓血细胞的自动识别,构建了骨髓红系细胞和粒系细胞数据集,基于深度学习语义分割技术提出了Cell Net网络模型。该模型通过加入残差模块增加了网络的深度,利用卷积残差块使网络模型更容易训练,并结合U-Net的裁剪操作为分割提供更精细的特征。实验结果表明,该模型对骨髓红系细胞和粒系细胞识别正确率分别达到93.65%、95.25%,为骨髓血细胞自动识别技术提供了一种方法。In order to realize the automatic identification of bone marrow blood cells,bone marrow erythroid and granulocyte data sets are constructed,and a CellNet network model is proposed based on deep learning semantic segmentation technology. The model increases the depth of the network by adding a residual module,uses a convolution residual block to make the network model easier to train,and combines the U-Net clipping operation to provide more refined features for segmentation. The experimental results show that the correct recognition rate of this model for bone marrow erythroid cells and granulocytes reaches 93. 65% and 95. 25%,respectively,which provides a method for automatic identification of bone marrow blood cells.

关 键 词：骨髓血细胞 细胞形态学 细胞分类 图像识别 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]