基于多维散射光检测技术的白细胞亚类识别方法的研究

Blood analysis system at present is complex in structure, high in price, high in maintenance and use cost, and poor in accuracy, consistency and comparability of detection results. Aiming at these flaws, a novel approach will be explored for the classification of leukocytes subgroup and cell recognition. The scattering light detection optical system designed by the applicant and his research group, which differs from current mainstream systems and has already awarded the invention patent license, will be adopted in this subject as a platform. Taking the classification and identification process of leukocytes subgroup as research object, this subject will make a thorough inquiry into the characteristics of cell signal, the form of noise under the novel detection method mode, and then explore the approach to obtain cell signal. Using the time-frequency analysis technology for nonlinear and non-stationary signal, the method to extract eigenvector of blood cells signal will be studied, especially the individual cases analyzing method for abnormal cell signal. Different on-line identification models which can describe different kinds of blood cells characteristics effectively will be constructed respectively using the intelligent modeling theory and method, cell signal identification algorithms matching with model will be investigated for the purpose of identifying blood cells signal steady, rapidly and accurately. According to clinical cases and experience of experts, expert system of blood cell detection and analysis oriented to disease diagnosis will be established and corresponding algorithm will be provided for the retrieval of rules generated by expert system. The research results of this project are in favor of simplifying the structure and enhancing the identification precision of blood analysis system, and consequently promoting and improving the diagnosis level in clinical and treatment of our country.

针对目前血液分析系统结构复杂，价格昂贵，使用和维修成本高，检测结果准确性、一致性和可比性较差的缺陷，探索新的白细胞亚群分类方法和细胞识别方法。课题以申请人及团队设计并已获发明专利授权的，不同于目前主流血细胞分析系统的多维散射光检测系统为平台，以白细胞亚群分类和识别过程为研究对象，研究在新的检测方法模式下细胞信号的特征和噪声的成因，探索细胞信号的获取方法；利用非线性、非平稳信号的时频分析技术，研究提取血细胞信号特征向量的方法，突出对异常细胞信号的个案分析；利用智能建模理论和方法，分类构建有效描述血细胞特征的在线识别模型，研究提出基于模型的细胞信号识别算法，实现稳定、快速、准确识别血细胞信号；根据临床案例并结合专家经验，建立面向疾病诊断的血细胞检测分析专家系统，研究匹配的专家规则检索算法。本项目的研究，有利于简化血细胞分析系统结构，提高识别精度，促进和改善我国各类医院临床诊断和治疗。

目前主流的白细胞亚分类检测需结合激光散射、射频、化学染色等技术，结构复杂，实现、使用成本高。本项目以多维散射光检测系统为平台，以白细胞亚群分类和识别过程为研究对象，研究构建了有效描述血细胞特征的在线识别模型，实现了血细胞信号稳定、准确、快速的分类识别。.项目验证了多维激光散射检测系统的有效性，并优化了多维散射光学结构，将后向散射光强度提高了一倍，有效滤除了偏振光，降低了背景杂散光，提高了检测结果的准确性。.通过分析细胞信号的有效性，研究了血细胞信号的获取方法，实现了对背景噪声的局部消除，并成功研制出血细胞识别电路和血液分析仪的微处理器/现场可编程门阵列两极控制系统。.本项目探索了能自适应地分解非线性非平稳信号的希尔伯特-黄变换（HHT）算法在白细胞信号（WBS）时频分析和分类中的应用效果，发现了各类白细胞信号在平均强度、谱质心以及能量贡献率上具有良好的区分度，并将其提取作为细胞分类的特征向量。.以WBS的特征向量作为输入序列，采用支持向量机分类器构建了健康人与患者的识别模型，识别率达到94.83%；采用RBF神经网络算法构建了白细胞亚类的分类模型，其分类结果与国外知名仪器Mythic 22测试结果相近，其中LYM、MON、GRAN分类相关性系数分别为：99.23%，94.33%，99.92%。.针对目前血液分析仪无法自动诊断血液疾病的现状，本项目以专家经验、临床复检规则和临床案例为知识源构建了产生式树形结构的知识库，并结合基于Snort高效匹配规则的推理机设计出了专家临床诊断系统。系统诊断的假阳性率为2.7%，假阴性率为3.31‰，满足临床诊断的相关指标。.本项目探索了新的白细胞亚群分类方法和细胞识别方法，为突破五分类血液分析仪的重大关键技术，研制出自主创新和具有国际竞争力的优质产品提供了理论和技术支持。