基于可视化模式识别技术外周血涂片成像的白细胞形态分类研究  

Study on the morphological classification of white blood cells by peripheral blood smear imaging on the basis of visual pattern recognition technique

作　　者：金婷 缪雪燕 徐燕 JIN Ting;MIAO Xue-yan;XU Yan

机构地区：[1]温州市中心医院采购中心,浙江温州325000 [2]温州市中心医院设备科,浙江温州325000 [3]温州市中心医院检验科,浙江温州325000

出　　处：《中国医学装备》2020年第1期2-5,共4页China Medical Equipment

摘　　要：目的:为有效降低医院检验人员的工作量,研究一种适用于外周血涂片成像的白细胞分类计数识别模型。方法:通过了解外周血细胞组成、白细胞分类计数和白细胞识别原理,分析可视化模式识别技术的白细胞增强和定位原理,引入白细胞自适应阈值分割技术,根据白细胞基本特征、细胞核形态特征以及细胞浆纹理对白细胞特征的选取,使用最小风险准则建立白细胞分类模型,使用临床1024幅血细胞图像作为训练集(512幅)和测试集(512幅),对白细胞分类计数模型进行训练和测试。结果:训练集模型白细胞识别率达96.7%,残差Loss值趋于在0.08;测试集嗜酸细胞、单核细胞、淋巴细胞、分叶核细胞、杆状核细胞和嗜碱细胞正确识别率分别为96.1%、97.1%、98.4%、89.9%、75.7%和95.6%。结论:基于可视化模式识别技术的白细胞分类计数模型具有较高的识别率,具有一定的鲁棒性和实用性,可有效降低检验人员的工作量。Objective: To propose a white blood cell classification and counting recognition model which was suitable for peripheral blood smear imaging so as to effectively reduce the workload of inspection staffs in hospital. Methods: By understanding the composition of peripheral blood cells, the classification and counting of white blood cells, and the principles of recognizing white blood cell, to analyze white blood cell enhancement and localization principle of visual pattern recognition technique, and to introduce self-adaptive threshold segmentation technique of white blood cell. Based on the basic characteristics of white blood cells, nuclear morphological characteristics, and the selection of cytoplasm texture for characteristics of white blood cells to use the minimum risk criterion to establish classification model of white blood cell. And clinical 1024 blood cell images were used as the training set(512 cases) and test set(512 cases) to train and test the classification and counting model of white blood cell. Results: The recognition rate for white blood cell in training set model could reach 96.7%, and the Loss value of the residual tended to be 0.08. And the correct recognition rates of the acidophilic granulocyte, monocyte, lymphocyte, segmented neutrophil, rod-shaped nucleus cell and basophil in test set were 96.1%, 97.1%, 98.4%, 89.9%, 75.7% and 95.6%, respectively. Conclusion: The classification and counting model of white blood cell based on visual pattern recognition technique has a higher recognition rate, and has certain robustness and practicability, and it can effectively reduce the workload of inspection staffs.

关 键 词：可视化模式识别 白细胞形态 形态分类 自适应阈值分割 最小风险准则 

分 类 号：R446.11[医药卫生—诊断学]