【作者】 刘伟；

【导师】 张崇巍； 王建平；

【作者基本信息】 合肥工业大学 ， 电气工程， 2013， 博士

【摘要】 系统动态的掌握个人身体机能,了解身体健康状况,树立健康意识是目前国家卫生健康领域的一个重要课题。生物电阻抗分析技术是一项无创、廉价、安全、无毒无害、操作简单的现代医疗检测方法,在人体健康信息的获取与监测中具有广阔的应用前景。体成分和内脏脂肪含量的改变是人体健康状态发生变化的重要诱因,也是普通人群易于分析的两个重要健康指标。传统的医疗检测手段操作复杂、检测成本较高,有的甚至会对人体带来一定伤害。应用生物电阻抗分析技术进行体成分、内脏脂肪含量的测量,并进行健康评估是实现个体健康状态监测和预警的有效手段。然而由于生物组织阻抗包含的生理状态信息并不充分,与人体体成分、内脏脂肪含量相关的关键参数较多,人体健康状态的科学分析过程中存在较多不确定性因素等问题,目前个人对健康状态的认识来源仍然是医学体检。如何利用人体生物电阻抗中带有的生理、病理信息,研究基于生物组织电特性和机器学习理论的体成分、内脏脂肪含量的测量模型并进行人体健康状态评价的新方法,具有重要意义。本课题的关键问题是在人群样本有限、阻抗信号资源有限、健康特征信息关系复杂条件下,通过科学的评价推理过程得到客观的结论。论文在对生物电阻抗技术理论研究的基础上,通过对人群样本相关数据的分析,得到体成分、内脏脂肪含量与样本个体参数的统计关系,利用机器学习领域的研究成果建立预测模型,构建了基于智能结构的健康状态云模型评价系统,主要研究内容如下：1.对人体生物电阻抗测量原理及方法进行了研究。包括人体生物电阻抗测量原理,生物组织电阻抗等效模型、频率特性,基于生物电阻抗的体成分和内脏脂肪面积相关的组织阻抗测量方法等问题,并对测量过程中影响结果的因素做出分析总结。2.对人体体成分预测模型进行了研究。通过样本人群数据的统计,分析了体成分与个体年龄、身高及性别的关系,针对阻抗测量过程中出现的噪声,设计了一种基于中值数绝对偏差的数据滤波器,消除了测量数据中由于突变性扰动或尖脉冲干扰导致的过失数据和随机噪声；将KLPS回归算法应用于人体TBW和FFM的预测,解决了特征数据可能存在的共线性和非线性；针对阻抗信息的局限性,采用相似性局部样本集进行训练学习,针对每个新样本单独建模并做出预测；通过定义训练样本集与待测样本的相似度,在对待测样本特征提取的同时进行KPLS回归；将KPLS、BP神经网络、多元回归、PLS等建模方法分别应用于体成分预测,对比了其在样本人群中的预测结果,通过与标准值的相关性、均方差分析,验证了本文提出的算法在人体成分预测中的有效性和优越性。3.对人体内脏脂肪面积的预测模型进行了研究。首先分析了与人体内脏脂肪含量相关的个体生理参数；针对基于生物电阻抗的人体内脏脂肪面积预测相关参数多,参数间存在一定相关性,训练样本较小,难以建立精确预测模型的问题,提出将支持向量机回归(SVR)应用于人体内脏脂肪面积的训练建模;针对建模过程中输入特征变量选择和SVR参数优化问题,提出将SVR和赤池信息准则(AIC)相组合的模型优化算法,将回归中的参数选择与支持向量机的参数优化看做一个组合问题,采用PSO算法进行搜索,实现SVM参数和输入变量的同时优化；对本文所提方法与KLPS回归、传统交叉验证SVR回归算法及多元回归方法在内脏脂肪面积预测中的结果进行了对比,通过相关性、均方差分析验证了本方法的有效性和优越性。4.对人体健康状态评价系统进行了研究。针对人体健康评价受影响因素多、人群样本非均质、相关数据庞大等问题,借鉴智能控制领域思想将人体健康评价系统分解为决策,测评两个层次,两层分别按智能计算、定性分析、定量执行三个阶段处理不同任务,构建了一种人体健康状态评价系统的智能结构模型,将研究目标不放在对象模式识别的数学模型上,而是建立被测目标模式特征模型的定性与定量结合的知识模型,解决了人体健康测评中任务复杂不易建模的问题；针对在健康评价过程中评价信息的模糊性和随机性,提出了基于自然语言的多属性群决策云模型集结方法；建立基于先验知识的健康评价指标属性与自然语言评语的标准云模型和基于实际测量指标的个人属性实际云模型；通过定义两种云模型的相似度,并进行计算得到测试者的健康评语；通过实例分析表明该评价方法在一定程度上客观的表达了自然语言在定性评价时的亦此亦彼,说明了本方法的可行性,为多属性评价问题提供了一个新思路。5.设计了一套用于人体生物电阻抗测量的仪器装置,完成了人体健康测评系统软件的编程。通过系统实验分别对测量仪器的重复性、相关性进行了分析;给出了详细的实验步骤、实验条件及实验结果,通过对结果的分析验证了设计仪器的有效性。仿真实验和实际应用结果均表明,本文所做的应用基础理论研究和给出的设计方法可适应于人体体成分、内脏脂肪的测量,结果的相关性、均方差优于其他方法,提出的基于云模型的智能结构评价系统适用于人体健康评估,课题的开展为国人健康管理领域健康促进的研究做出了探索。更多还原

【Abstract】 It is an important subject to monitor human body function, physical quality systematically and dynamical to improve human health consciousness in Chinese health management fields. Bioelectrical impedance analysis (BIA) technology is a modern medical detection method with low-cost, non-invasive, simple operation, wealth of information, which has a wide application prospect in the human health monitoring and evaluation fields.The change body Compositon and visceral fat is an important incentive to the human health change, as well as the two health indexes are also easy to analyze for normal population. It is high-cost, complex operation and even harmful of the traditional medical measurement methods. It is an effective method for human health monitoring and alarm using BIA to measure body Compositon and visceral fat to evaluate the body health status. While the health information in the bioelectrical impedance is insufficiency and the parameters related body compositon and visceral fat are uncertainty, which lead to the difficulty in the filed of human health status analysis. Now the medical examination is still a primary method for health recognition. The electrical characteristics of human biological tissue is change with the pump-frequency, but it always incudes the human physiology, pathology information. So it is significance to research a new method to predict the body compositon, visceral fat and evaluate the health status using BIA and machine learning theory.The key problem of this project is how to get the scientific conclusion through humanoid inference in the case of the limited human specimen, limited information of impedance signal resource, complex relationship of health feature. In this paper, the relationship between the individual physiological parameters and body Compositon, visceral fat is analyzed through collection of specimen data. The prediction model is built using the results of machine learning fields and the intelligent structure evolution system for health is designed using cloud model. The main research contents are as follow:1. The theory of human bioelectric impedance and the measure method are studied. It includes the model of human body biology tissue impedance equivalent model, the problem of response frequency of bio-issue, the measurement methods of bioelectrical impedance related body Compositon and visceral fat et al, and the factors influence results in the process of measurement are analyzed2. The prediction model for human body Compositon of Chinese is studied. The relationship among the body Compositon and the age, height, weight is analyzed through the specimen population data statistics. A new method data filter on improved median number absolute deviation is proposed to filter impedance data, which can eliminate the fault data and random noise brought by mutational perturbation or sharp pulse jamming; the KPLS algorithm is applied in the human TBW and FFM prediction, the colinearity and nonlinear of the feature data can be avoided; aiming to the limited information of impedance, the similarity of local sample set is chosen to train, the poetical model is obtained for the new sample of testing sample set; the similarity local sample set is obtained through definition of the similarity between the train set and the testing set; the modeling methods include KPLS, BPNN, multiple regression, PLS are studied and applied in the body compositon prediction. The better method is given through comparison of the results using above several methods to predict the body compositon. The modeling method is proven efficient through computing of the coefficient and standard error of the test sample.3. The prediction model for human visceral fat area is studied. The relationship among the visceral fat area and the age, related impedance,weight,abdomenal shape are analyzed through the specimen population data statistics; due to the complex signals and many interfering factors in the process of assessing the viscera fat area using bioelectrical impedance, it is hard to get an accurate predicting model, the Support Vector Regress (SVR) algorithm is applied in the visceral fat area prediction; the optimum of the SVR parameters and the selection of the feature parameters can be regarded as a compound optimum problem, a new optimization algorithm on SVR and Akaike information criterion (AIC) is presented; the improvement PSO optimal algorithm is used to search the optimal value of the objective function to improve the efficiency; in the experiment, KPLS, traditional SVR, multiple regression are applied in the visceral fat area prediction, the result shows the method presented in this paper has valuable and good capability to measure the human viscera fat area. 4. The evaluation system of human health status is built. Aiming to many interfering factors, the heterogeneous of human specimen and the huge data in the health evaluation process, the complex task can be decomposed into two levels:decision and evaluation according to the intelligent control thought, each level can be decomposed into three stage that intelligent computing, qualitative analysis and quantitative evaluation, thus the emphasis of the research is not on the mathematic model but on the knowledge model that is combined the qualitative judgment and quantitative calculation, which solve the model problem; aiming at the problem of uncertainty and the transform between the qualitative and quantitative in evaluation of human health process, the multi-attribute-evaluation model based on cloud model is presented; the multi-attribute-evaluation individal preference cloud model aggregation method based on natural language is given; the standard evaluation cloud model on health evaluation index and natural language is built; the practical cloud model is obtained through the tester’s health index, the conclusion of tester’s health status is drew through definition of similarity. Through the example analysis, the method presented in paper can show the obscurity of the nature langue objectively and the method is feasible, which bring about a new thought of the multi-attribute-evaluation problem.5. The bio-impedance measurement instrument for human body is designed, the software of human health evaluation is completed. The repetition and correlation of the instrument are analyzed through experiments; the detail procedure, condition and results of experiments are given; the conclusion of the results shows valuable of the instrument.The simulation experiments and the practical application show that the instrument designed this paper can be used in practice, the research of application fundamental theory in this paper and the presented methods can be used to predict body Compositon and visceral fat area, the correlativity and standard error are better than other models, the intelligent structure evaluation system on cloud model is suitable for health evaluation, the subject makes exploration in the field of Chinese health management.更多还原