基于视觉感知融合的中医诊断分析模型研究

发布时间：2018-05-05 15:47

本文选题：视觉感知 + 诊断特征　；参考：《哈尔滨工业大学》2017年硕士论文

【摘要】：中医诊断的辨证论治以望、闻、问、切四诊并用为依据,而望诊就是通过观察病人的神、色、形、态等体外诊察信息来获取病人的视觉感知特征。为了克服传统望诊依赖于医师主观诊断的局限性,并有效解决中医望诊现代化研究中面临的可复用性问题。本课题的研究从望诊的客观化入手,利用标准化视觉感知采集设备,在合作医院采集了舌象、面部及舌下静脉三类视觉感知图像,并利用数字图像处理技术进行了量化处理,提取出基于统计特征与病理特征的7种诊断特征用于分析研究,诊断特征包括舌象的颜色特征,舌象的几何特征,舌象的纹理特征,面部的颜色特征,面部的纹理特征,舌下静脉的颜色特征,舌下静脉的几何特征。基于大规模视觉感知特征,本课题重点研究了诊察数据、量化特征和疾病的复杂关联关系。本课题利用模式识别方法,通过对不同健康状态下的被诊察者提取的多种感知特征进行融合分析,探索了隐藏于多种诊断特征背后的规则和规律,为辅助临床诊断提供了有力支持。本课题主要涉及的诊断融合策略有特征级融合与决策级融合两种。特征级融合通过对某种疾病相互独立的多个子特征的分析,寻找出该疾病相对于健康状态下表现性较强的一组特征,并利用典型关联分析理论将其融合成一个新的特征,进而通过分析这个新特征来做出最终的诊断判读。同时,本课题也使用了基于核化的典型关联分析来融合一组诊断特征,以便进一步分析诊察特征间的关系。决策级融合通过对多特征、多分类器的局部识别结果的相辅相成,将这些局部识别结果在一定数学模型的基础上进行融合以获得模式样本的最终输出结果,进而提高整个识别系统的性能。本课题针对不同的分类器输出,将基于贝叶斯理论的融合方法引入到多诊断特征分析中,并使用基于表决规则的融合方法与其做性能对比;随后通过对SRC与SVM两个分类器的性能分析,提出了一种基于SRC与SVM的联合分类器,用以改善单分类器的识别结果。最终的健康与疾病二分类的实验结果表明,不同的融合方法对于视觉感知特征的诊断分析来说,都能发挥出一定的作用,这表明感知融合计算对于中医望诊的现代化研究有着重要的意义。
[Abstract]:The diagnosis and treatment of TCM is based on looking, smelling, asking and cutting four diagnoses, and the diagnosis of hope is to obtain the patient's visual perception characteristics by observing the patient's mental, color, shape and state in vitro. In order to overcome the limitation of traditional inspection relying on the subjective diagnosis of doctors, and to solve the problem of reusability in the modern research of TCM inspection. The research of this subject starts with the objectification of inspection, using standardized visual perception acquisition equipment, collects three kinds of visual perception images in cooperative hospital, such as tongue image, facial image and sublingual vein, and makes use of digital image processing technology to carry out quantitative processing. Seven diagnostic features based on statistical and pathological features are extracted for analysis and research. The diagnostic features include tongue color, tongue geometry, tongue texture, facial color, facial texture, and so on. The color characteristics of the sublingual vein and the geometric characteristics of the sublingual vein. Based on the characteristics of large-scale visual perception, this paper focuses on the complex relationship between diagnostic data, quantitative features and diseases. In this paper, the pattern recognition method is used to study the rules and rules hidden behind the various diagnostic features through the fusion and analysis of the various perceptual features extracted by the examinees under different health conditions. It provides support for clinical diagnosis. There are two kinds of diagnostic fusion strategies: feature level fusion and decision level fusion. By analyzing several independent sub-features of a disease, the feature level fusion finds out a set of features that are more expressive than that of a healthy state, and uses the theory of canonical correlation analysis to fuse the disease into a new feature. Then, the final diagnostic interpretation is made by analyzing this new feature. At the same time, a typical correlation analysis based on kernelization is used to fuse a set of diagnostic features in order to further analyze the relationship between the features. Decision level fusion complements the local recognition results of multi-feature and multi-classifier, and fuses these local recognition results on the basis of certain mathematical models to obtain the final output of pattern samples. Then the performance of the whole recognition system is improved. According to different classifier output, this paper introduces Bayesian theory based fusion method into multi-diagnosis feature analysis, and compares its performance with the fusion method based on voting rules. Then, by analyzing the performance of SRC and SVM classifiers, a joint classifier based on SRC and SVM is proposed to improve the recognition results of single classifier. The results of the final two classification of health and disease show that different fusion methods can play a certain role in the diagnosis and analysis of visual perception characteristics. This indicates that perception fusion computing is of great significance to the modern research of traditional Chinese medicine (TCM) diagnosis.
【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】：2017
【分类号】：R241;TP391.41

【相似文献】