摘要
目的:开发并应用血液白细胞图像自动识别系统。方法:运用数据挖掘、模式识别和人工智能等技术,并与手工显微镜分类计数法(参考方法)比较。结果:白细胞图像自动识别系统的准确率为94.9%,重复率为0.1%,漏检率为0.6%,稳定性达到97.6%,检测速度低于5min/片。与手工显微镜分类法的相关系数:中性粒细胞为0.997 1,淋巴细胞为0.994 2,嗜酸性粒细胞为0.989 9,单核细胞的相关系数为0.982 2,嗜碱性粒细胞为0.852 8,其他细胞为0.979 8。结论:研制的血液白细胞图像自动识别系统的检测结果与手工显微镜分类计数法相近,能够满足临床要求。
Objective： To develop and apply automatic recognition system of leukocyte images. Methods： Data mining, pattern recognition, artificial intelligence and other techniques were applied to design this new system. Its accuracy, repeatability error , omission error, reliability, detection rate and correlation coefficients were evaluated according to manual differential counts （reference method）. Results： The accuracy is 94.9 % ; repeatability error is 0.1% ;omission error is 0.6%; reliability is 97.6%and detection rate is below 5 minutes per slide＇respectively. The correlation coefficients are the highest for neutrophils （0. 997 1）, lymphocytes （0. 994 2） and eosinophils （0. 989 9）. For monocytes, the correlation coefficient is 0. 982 2. The lowest correlations are hasophils （0. 852 8） and other cells（0. 979 8）. Conclusion： The evaluation shows this system has performance at least equal to that of direct clinical microscopy and could meet the clinical demand.
出处
《实验技术与管理》
CAS
北大核心
2012年第12期44-47,50,共5页
Experimental Technology and Management
基金
广东省科技支撑计划资助项目(2010A030500013)
关键词
白细胞
图像分析
自动识别
性能评价
leukocyte differential
image analysis
automatic recognition
performance evolution
