舌下静脉血管分割及其病理特征分析研究

发布时间：2018-10-11 19:04

【摘要】：中医望诊是一种传统的疾病诊断方式,其中舌像诊断是很重要的一个组成部分。舌像诊断主要分为舌面诊断和舌下静脉诊断等,对于舌面诊断已经有了许多针对性的研究,而且经过多年的发展已经产生了许多成果,但是关于舌下静脉的研究还比较的少,有待进一步的探讨。本课题主要是在舌下静脉的影像信息中,寻找对于疾病分析有潜在价值、有指导意义的特征,并根据这些特征来进行健康与疾病的分析。具体研究内容包括:图像的采集与预处理、舌下静脉血管的分割、特征选择与优化以及基于舌下静脉血管特征的聚类分析。通过与医院合作,采集到舌下静脉图像的样本,同时得到对应样本的各类生化指标以及医生所做出的健康状况判定结果,并把这些信息作为该样本的疾病标签。截止目前已采集含标签样本两千多例。本课题选取其中采集样本较多的类别,如肺癌、高血压、乳腺癌、肾病、失眠、糖尿病、胃炎、肿瘤以及健康样本等作为数据集。对于采集到的原始图像采用一种基于多项式的校正算法进行颜色校正,并提出了一种基于HSI与LUV颜色空间像素生长法的交互式分割算法进行舌下静脉图像的分割。基于1000幅分割后的舌下静脉图像构建了舌下静脉的颜色空间,并通过k-means聚类方法获得了舌下静脉基于该颜色空间的颜色特征;同时,根据分割的舌下静脉提取了基于RGB和HSV颜色空间下的颜色特征,以及舌下静脉长度、宽度以及长宽比等几何特征,并通过不同的特征组合进行了特征向量的优化。利用舌下静脉特征向量对采集的图像样本进行了聚类分析,主要进行了健康与几种疾病的二分类,通过不同的二分类算法的比较选择了SVM分类器算法进行二分类,最后分类结果的平均正确度为80.88%,而对于几种典型的疾病如2型糖尿病与健康的分类精度可以达到89.34%,这说明舌下静脉这一特征对于疾病分析确实有意义。实现了SVM决策树的多分类器,对于健康、失眠和乳腺癌三类的分类准确度可以达到70.19%,说明SVM决策树算法对于多分类问题可以有比较好的效果,并且证明舌下静脉对于疾病的诊断分析是有意义的。
[Abstract]:Traditional Chinese medicine diagnosis is a traditional way of disease diagnosis, among which tongue image diagnosis is an important part. Tongue image diagnosis is mainly divided into tongue surface diagnosis and sublingual vein diagnosis. There have been many targeted studies on tongue surface diagnosis, and after years of development, many achievements have been produced, but the research on sublingual vein is still relatively few. Further discussion is needed. In this paper, we mainly look for the potential and instructive features of disease analysis in the image information of sublingual vein, and analyze the health and disease according to these characteristics. The specific research contents include: image acquisition and preprocessing, sublingual vein segmentation, feature selection and optimization, and clustering analysis based on sublingual vein features. By cooperating with the hospital, the samples of the sublingual vein images were collected, and the biochemical indexes of the corresponding samples and the results of the doctor's health assessment were obtained, and the information was used as the disease label of the sample. Up to now, more than 2,000 samples containing labels have been collected. In this study, we selected a large number of samples, such as lung cancer, hypertension, breast cancer, nephropathy, insomnia, diabetes, gastritis, tumor and health samples as data sets. A polynomial correction algorithm is used to correct the original image and an interactive segmentation algorithm based on HSI and LUV color space pixel growth method is proposed to segment the sublingual vein image. The color space of the sublingual vein is constructed based on 1000 segmented images of the sublingual vein, and the color feature of the sublingual vein based on the color space is obtained by k-means clustering. Based on the segmented sublingual veins, the color features based on RGB and HSV color spaces, as well as the geometric features such as the length, width and aspect ratio of the sublingual veins are extracted, and the feature vectors are optimized by different feature combinations. The sublingual vein feature vector is used to cluster the collected image samples. The two classification methods of health and several diseases are mainly carried out. The SVM classifier algorithm is selected to do the two classification by comparing different two classification algorithms. The average accuracy of the classification results is 80.88, while the classification accuracy of several typical diseases such as type 2 diabetes and health can reach 89.34, which indicates that the sublingual vein is of great significance for disease analysis. The classification accuracy of SVM decision tree can reach 70.19 for health, insomnia and breast cancer. It shows that the SVM decision tree algorithm has good effect on multi-classification problem. It also proves that the sublingual vein is of significance in the diagnosis and analysis of the disease.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：R241;TP391.41

【相似文献】