基于机器学习的白细胞六分类研究
2020-12-23孙凯姚旭峰黄钢
孙凯 姚旭峰 黄钢



摘 要: 鉴于现有机器学习白细胞分类方法难以验证网络泛化性与鲁棒性,本研究提出了一种联合多类型特征的白细胞分类方法。首先使用图像剪裁与中心化及改进型颜色阈值分割完成11865张六种类别白细胞图像的预处理与分割操作。随后在特征提取部分筛选出细胞几何、纹理、小波三部分共63个特征;通过主成分分析法实现了降维后得到8个主成分。最后使用支持向量机、多层感知机与决策树分别进行分类工作,结果为高质量图像最高88.6%;噪声图像最高84.5%;低分辨率图像最高87.6%的分类精度。实验结果验证了所提出方法的鲁棒性和泛化性好,可实现白细胞的准确分类。
关键词: 机器学习;图像处理;分类;白细胞
中图分类号: TP391.41 文献标识码: A DOI:10.3969/j.issn.1003-6970.2020.10.024
本文著录格式:孙凯,姚旭峰,黄钢. 基于机器学习的白细胞六分类研究[J]. 软件,2020,41(10):98101+134
【Abstract】: In view of the difficulty in verifying the generalization and robustness of existing machine learning leukocyte classification methods, this study proposes a white blood cell classification method that combines multiple types of features. First, use image cropping and centralization and improved color threshold segmentation to complete the preprocessing and segmentation of 11,865 six subtypes of white blood cell images. In the feature extraction part, 63 features including cell geometry, texture and wavelet were screened out; 8 principal components were obtained after dimensionality reduction was achieved by principal component analysis. Finally, support vector machines, multi-layer perception and decision trees were used for classification. The results were high-quality images up to 88.6%; noise images up to 84.5%; low-resolution images up to 87.6%. The experimental results verify that the proposed method has good robustness and generalization, and can achieve accurate classification of white blood cells.
【Key words】: Machine learning; Image processing; Classification; White blood cells
0 引言
血液白細胞通常可以划分为五种类型,分别为嗜碱性粒细胞(Basophils,bas)、嗜中性粒细胞(Neutrophilic,neu)、单核细胞(Monocytes,mon)、淋巴细胞(Lymphocytes,lym)以及嗜酸性粒细胞(Eosinophils,eos),嗜中性粒细胞可分为嗜中性杆状核粒细胞(Band Neutrophil,ban)以及嗜中性分叶核粒细胞(Segmented Neutrophil,seg)[1]。白细胞根据形态差异又可以分为颗粒和无颗粒两类细胞,其中嗜酸性粒细胞、嗜碱性粒细胞及中性粒细胞属于颗粒细胞,单核细胞及淋巴细胞属于无颗粒细胞[2]。白细胞浓度异常可能代表某些血液疾病的发病迹象[3]。因此,实现白细胞的精确分类有助于帮助医生筛查白细胞相关病症,在临床应用中具有重要研究意义[4-5]。……
