基于BP神经网络的血液细胞自动识别技术的研究与实现

【作者】 郑辉；

【导师】 陈曙；

【作者基本信息】 山东大学 ， 电路与系统， 2007， 硕士

【摘要】 随着科学技术的发展，计算机技术在医学领域得到了广泛的应用，其中计算机辅助医疗外科技术及辅助诊断技术越来越受到人们的青睐，尤其是在医学显微图像处理方面的应用更是倍受关注。传统的医学显微检测工作效率低、强度大，存在主观的观察误差，对检测员的技术水平要求较高；并且它只能通过显微摄影进行图像记录，不能对图像进行必要的处理，也不能快速存储和通过网络远距离传输图像信息，因而已经无法适应现代医学显微图像发展的要求。目前，随着光电转换技术的日臻成熟、计算机硬件的高速发展和功能强大的图像处理软件的出现，实现彩色数字化医学显微图像实时显示的条件已经具备。根据当前血液细胞观测的新需求，现代医学发展趋势迫切需要我们把传统技术难以观测到的显微镜视场光学图像显示在计算机屏幕上，并转换成大面积、高亮度、真彩色能被自动识别的图像，以减轻专业技术人员在显微镜下从成千上万个目标中寻找、观测、捕捉异常小个体的劳动强度，然后对观测细胞的色彩、形态等参数进行定量地分析统计，辅助医生进行医疗诊断。本文的研究工作主要围绕着基于BP神经网络的数字图像处理技术展开的，重点探讨了BP神经网络用于血液细胞识别的可行性，并且使用VC++编程实现了细胞自动识别计数系统。论文首先介绍了人工神经网络的基本概念，对人工神经网络的构造模型进行详细的阐述，列举了几种常用的神经网络模型，重点对BP神经网络做了全面的讨论；接着介绍了数字图像处理的基本理论如图像平滑、直方图阈值分割、细化、梯度修正以及数学形态学等，及这些理论在细胞识别方面的具体应用和处理的结果；然后提取细胞图像的特征，并根据这些特征利用BP神经网络进行细胞分类，最终实现各种细胞的自动计数，从而为医学诊断提供了客观依据；结尾对本文的研究工作做了总结，并对后续研究工作提出了建议。本文的研究工作为医学诊断学提供了一种有效的分析方法，在提高细胞计数的效率和准确率方面有较高的实用价值，也为将来更多种类细胞的识别分类提供了参考。更多还原

【Abstract】 With the development of computer technology, computers are widely used in medical study field. Among all these applications, computer assisted diagnosis are getting more and more attention, especially the processing of medical microscopic images. The work is very tiring with low efficiency in the traditional methods. So the members of the examination often have subjective observation errors and are requested to have higher technique level. Images can only be recorded by microscopic photograph however can not be processed, saved, and delivered long-distance by the network, so it isn’t fit for the new request of development. Now we can get the digital medical microscopic images with the modern light-electricity conversion, hardware of the computer, data and image processing software etc. To reduce work intension and assist diagnosis for doctor, the blood cells need to be recognized by the specific request and analyzed on the quantitative colors and morphology in the computer.The paper is based on BP neural network method in digital image processing. Based on the method, the cells auto-counting system is developed by VC++. Firstly, we introduce the basic concepts and expound several familiar models about artificial neural network in detail, then make overall discussion on the BP neural network. Secondly, we introduce basic theories of digital image processing such as image smoothing, histogram threshold segmentation, thinning, gradient enhancement, mathematical morphology. The application of these knowledge and its results are also introduced. Thirdly, we use the BP neural network to classify the kind of cells with the extracted characteristic, and count the number of cells to give proof for medical diagnosis. At Last, we review the entire work and suggest the direction for future research.This paper provides theoretical proof for medical diagnosis and improves efficiency and accuracy for identifying more kinds of cells.更多还原