【作者】 黄勃；

【导师】 张大鹏； 王宽全；

【作者基本信息】 哈尔滨工业大学 ， 计算机应用技术， 2009， 博士

【摘要】 中医学是一种起源于中国,以古代中国人的医学实践为主体的传统医学。检查舌象的状态是中医诊断学中应用最广泛、最有价值的方法之一。舌诊利用各种舌象特征,包括颜色、纹理、形状等,来进行诊断。在舌诊学中,舌在颜色、纹理、形状等方面的特征,既是内部病变的一种敏感标志,又是不可替代的一种区分中医证候的准则。近年来,许多学者致力于将模式识别与图像处理技术相结合,寻求中医舌诊不量化问题的解决方案。然而,由于中医舌诊过分依赖于个体的感受和经验,阻碍了这个学科的进一步发展。为了解决上述问题,本论文对中医舌诊现代化过程中舌图像特征分析与诊断分类等内容进行了初步的研究,主要包括：(1)舌色特征分析：颜色是一种视觉感知的属性,人体组织的颜色信息在医学诊断中起着重要的作用。在舌诊学中,舌色是用来进行诊断的最重要的依据。然而在诊断过程中,由于临床医生主观意识的影响,舌色分析的结果存在着大量不精确性和不确定性。为了尽量减少这些主观因素的影响,本文提出一个非参数化半监督学习的框架。它应用前向选择和后向选择来获得正确的像素样本,并用这些样本构建像素原型集。该像素原型集则可作为舌色量化的依据。实验结果显示：构建的像素原型集,可以用来将舌图像中的所有像素区分为12个舌色类别。(2)一种舌象纹理特征的分析(蕈状乳头增生识别)：本文提出了一种计算机纹理识别技术,来判定舌面上的蕈状乳头是否增生。首先进行纹理块获取,在经过检测反光点与去除反光点的预处理后,利用一组2维Gabor滤波器组,来提取和表达纹理特征,然后运用线性判别分析对舌图像数据库中的数据进行分类,实验结果合理的证明了方法的有效性。(3)舌形特征分析：本文提出一个舌形自动分类方法。首先利用三个几何准则来校正舌形歪斜；然后提出七个与长度、面积和角度相关的几何特征；接着用七个层次分析法的模块,将高度主观和抽象的专家评价解析成可度量的机器表达,每一个模块被用来决定一个舌图像是否属于一个特定的舌形类别；最终应用一个模糊融合的框架,来确立量化特征与舌形类别之间的不确定性。在362个样本的实验中,取得了90.3%的分类准确率。(4)八纲证候的诊断分类：该部分主要包括基于舌色冷暖属性判别的“寒证”“热证”分类。冷色、暖色、过渡色这些词也是舌色属性的一部分,这种舌色的冷暖属性与证候有着紧密的联系。本文提出了一个半监督学习(cluster and label)的模型,通过舌色将舌图像上的所有像素区分为3类：冷、暖或过渡色。这个模型首先利用期望最大化算法,将所有的像素区分为很多个聚类,这次聚类的数量较大(共150个聚类),而每个聚类的尺度较小。然后人工标定将赋予这些聚类3种舌色冷暖属性的类别标签：冷色、暖色、过渡色。最后使用一个查找表将所有像素区分为舌色冷暖属性的三个类别,并以此为依据,将舌图像样本区分为“寒证”或“热证”。除此之外,我们还利用前面章节所提取颜色和纹理特征作为特征矢量,进行诊断分类研究,列出了一些针对“虚证”“实证”分类、“表证”“里证”分类的初步研究结果,从而最终初步实现了对中医中最重要的辨证方法——八纲辨证的分类研究。本论文在舌特征提取和诊断分类方面进行了有益的、探索性的尝试,为中医舌诊自动化诊断提供了一种有效的解决途径。更多还原

【Abstract】 Traditional Chinese Medicine (TCM) includes a range of traditional medical practices originated from China. Examination of the condition of the tongue is one of the most valuable and widely used diagnostic methods in TCM diagnostics and takes account of a wide variety of features including color, texture, and shape. In tongue diagnostics, changes in color, texture, and shape of tongue are sensitive indicators of internal pathological symptom and indispensable guideline for overall health state. In recent years, many researchers are dedicated to the combination of modern pattern recognition and image processing technologies, trying to find solutions to the non-quantitative issue of traditional tongue diagnosis. However, the further development of traditional tongue diagnosis is limited by its dependence on individual visual sensation and experience.In this dissertation of computerized tongue diagnosis, we investigate tongue feature analysis and diagnosis classification, including:(1) Feature Extraction of Tongue Color:Color is a visual perceptual property, and chromatic information of human tissue plays an important role for medical diagnosis. Tongue color is the most important characteristic for identifying diseases in tongue diagnostics. However, because of detailed visual discrimination based on the experience and knowledge of practitioners, there is much uncertainty and imprecision associated with tongue color in medicine. In order to eliminate these subjective factors, we have established a nonparametric semi-supervised (cluster and label) scheme to obtain labeled pixel samples. It applies both forward and backward selection procedures for constructing the pixel prototype sets with class labels. These pixel prototypes can classify all pixels of a tongue image into twelve classes of tongue colors.(2) Feature Extraction of Fungiform Papillae Hyperplasia (FPH):We propose a computer-aided system for identifying the presence or absence of Fungiform Papillae Hyperplasia (FPH). We first define and partition a region of interest (ROI) for texture acquisition. After preprocessing for detection and removal of reflective points, a set of 2D Gabor filter banks is used to extract and represent textural features. Then, we apply the Linear Discriminant Analysis (LDA) to identify the data sets from the tongue image database. The experimental results reasonably demonstrate the effectiveness of the method described in this paper.(3) Feature Extraction of Tongue Shape:We present a novel approach to the automated classification of the tongue shape. The first step is to permit accurate positioning for analysis by applying three geometric criteria to correct tongue deflection. Then we developed seven geometric features using various measures of length, area, and angle on the tongue. And seven modules of Analytic Hierarchy Process (AHP) are constructed to decipher the highly subjective and metaphorical human judgments as a well-defined and measurable machine representation. Each one is utilized to decide whether a tongue image belong to a specified class of the tongue shape. Finally, to allow the reliable machine-classification of tongue shapes, we applied a fuzzy fusion framework to formalize the uncertainty between the quantitative features and tongue shape classes. In experiments conducted on a total of 362 tongue samples, our system achieved an accuracy of 90.3%.(4) Classification of Eight Principal Symptom:Major research of diagnosis modeling is the classification model of "cold" and "hot" symptom based on the chromatic characteristic of "warm or cool". The terms "warm", "neutral", and "cool" are used to refer to the color of the tongue and are associated with various states of health. We propose a semi-supervised (cluster and label) scheme for classifying all pixels of a tongue image by color into three categories:"warm" "neutral", or "cool". The proposed scheme makes use of a classical clustering algorithm, Expectation Maximization, to divide all pixels into 150 clusters. Here, for all pixels in tongue color gamut,150 is a relatively large number and each cluster is on a small scale. Then manual labeling endows these clusters as category labels. Finally, we use a lookup table to classify all pixels into three categories of "warm or cool". Based on this result, we classify these images into "hot" or "cold" symptom. Except that, we utlize some features extracted from aforementionedsection, to assist the diagnosis of some symptoms. In this section, we list some classification results of "excess" and "deficiency" symptom, "exterior" and "interior" symptom, and these results are very helpful for the most import diagnostic method in TCM, "eight principal symptom diagnosis"In this dissertation, we carry out some investigations on tongue feature extraction and diagnosis modeling, which will be helpful for the computerized tongue diagnosis.更多还原