【作者】 高洪智；

【导师】 卢启鹏；

【作者基本信息】 中国科学院研究生院（长春光学精密机械与物理研究所） ， 光学工程， 2011， 博士

【摘要】 血液的生化检测是健康诊断的重要手段。常规的抽血检验方法虽然能保证检测精度,但是存在以下不足:(1)有创伤;(2)检测周期长;(3)需用试剂。利用近红外光谱分析技术实现无创生化检测具有无痛苦、无感染、无需试剂、便于实现生化指标的实时监测等优点。但近红外无创生化检测实现起来却相当复杂,目前面临的难题主要有:(1)血液生化成分的信号微弱;(2)人体组织产生的强背景干扰。采用不同血流容积光谱相减的方法理论上能够消除人体组织产生的强背景干扰,得到纯净的血液光谱,因为在短时间(秒量级)内,人体的生理状态不变,包括人体组织结构、成分及血液的生化指标等都没有变,而血流容积却在脉搏搏动或血流阻断等自然或人为作用下发生变化。在这一短时间内,连续在人体某一部位测得一系列血流容积变化的近红外光谱,取不同血流容积的两幅人体光谱相减,就可以扣除人体组织产生的强背景干扰,得到一定血流容积的血液光谱。血流容积光谱相减法的目的是要对不含人体背景干扰的纯净血液光谱与其各生化成分的浓度信息建立多元校正模型,对未知血液生化指标的光谱进行预测,最终实现无创生化检测。然而由于自然或人为作用,血流容积发生变化时测得的人体光谱中所反映的血液光程是变化的,两幅人体光谱相减后得到的血液光谱的光程将是未知的,如果将各种未知光程的血液光谱直接建立多元校正模型,将会使模型的性能降低,这给血流容积光谱相减法带来了困难。需要注意的是,这些不同光程的血液光谱中生化成分的信息是不变的,变化的仅仅是光程信息。能否将各种不同光程的成分含量不变的血液光谱归一到一个不含光程信息的等效光谱上,也就是这些不同光程的血液光谱在反映血液生化成分信息的作用上是否是等效的,以及如何消除这些光程信息的影响并建立多元校正模型,这些问题是本论文所要研究的重要内容,也是不同血流容积光谱相减法能否得到实际应用的关键。本文围绕不同血流容积光谱相减法在实际应用中的关键问题展开了研究,包括不同光程的血液光谱等效性研究、不同光程光谱的多元校正模型研究、不同血流容积光谱相减性能的提升方法研究这三个方面。具体研究内容和主要结论如下:1)利用组织光学知识和化学计量学方法对不同光程光谱的等效性进行了研究。首先对不同血流容积光谱相减法的原理进行了推导,从原理上分析了不同光程血液光谱产生的原因及获取等效光谱的意义。对非散射介质中的不同光程的光谱等效性进行了研究,采用化学计量学方法能够得到归一化的等效光谱。通过对组织光学中光传输模型及修正的朗伯比尔定律的研究,认为不同光程的血液光谱同样是等效的。利用葡萄糖溶液和脂肪乳溶液的模拟实验对不同光程的光谱等效性进行了验证,并得到了各自的等效光谱。2)针对不同光程的近红外光谱在建立多元校正模型中遇到的问题,提出采用正交信号校正和各种校正空间的净信号分析方法用于提高多元校正模型的性能。通过对生物分子摩尔吸光系数建立的模拟光谱及各种葡萄糖模拟溶液进行分析,光程等干扰信息得到了抑制,所建立的多元校正模型性能得到了提高。3)通过模拟实验对不同血流容积光谱相减法在实际应用中的效果进行分析。采用光谱相减后同一光程光谱的欧氏距离平均值、标准差等指标来评价光谱相减的效果。针对光谱相减中的噪声干扰问题,分别采用Savitzky-Galay平滑滤波、小波变换滤波、经验模态分解滤波等滤波算法对噪声进行抑制,提升光谱相减的性能。经过实验验证,小波变换滤波、经验模态分解滤波对光谱相减的性能提升较为显著。本文的主要研究内容为血流容积光谱相减法在无创生化检测中的实际应用提供了理论和实验基础。更多还原

【Abstract】 Blood biochemical sensing has been a significant method in clinical diagnostic field. The effectiveness of conventional blood testing methods, which ensure a high accuracy, has been hindered by the following defects: (1) invasion; (2) long detection period; (3) requirement of reagents. As a promising noninvasive biochemical monitoring method, near infrared absorption spectroscopy analytic technique has many obvious advantages. The testing is painless, reagentless, and has no risk of infections. Besides, the method makes it convenient to process the real-time monitoring of blood biochemical indexes. However, the realization of NIR noninvasive biochemical sensing is rather complicated. Some major obstacles are: (1) weak signal of blood biochemical components; (2) strong interference of human tissue background.Theoretically, the spectral subtraction approach with different blood volume can eliminate human tissue background interference and obtain effective spectrum information of blood. This is because in an extremely short time period (seconds), human physiological states, including human tissue background and blood biochemical indexes, barely change. Yet as the result of pulsation or vascular occlusion, blood volume is all the time changing. Therefore, by subtracting two blood spectra measured in a row within this short period, information of certain-volume blood spectrum can be obtained, with human tissue background interference eliminated.In order to realize noninvasive biochemical sensing, the spectral subtraction approach with different blood volume aims at establishing calibration models to predict the biochemical indexes of unknown blood samples. However, the obtained spectra of different blood volumes actually reflect different blood pathlengths. So by subtracting the two spectra, we obtain a blood spectrum with the pathlength unknown. In this case, the model will be insufficiently effective if established directly with pathlength-unknown spectra, which is a big problem of subtracted blood volume spectrometry. Notably, the information of blood biochemical components remains constant despite of the changing pathlength. In this paper, whether the spectral subtraction approach with different blood volume can be practical applied is discussed. To look into this point, the paper involves several key problems—whether it’s possible to normalize spectra, same in concentration but different in pathlength, to an equivalent spectrum with no pathlength information; whether spectra with different pathlength information perform equivalently on reflecting information of blood biochemical components; how to eliminate the interference of the pathlength variations and thus establishing multivariate calibration models.The paper focuses on the key problems of the approach application of spectral subtraction with different blood volume. Particularly, it involves the research of the equivalence of blood spectra with different pathlengths. Besides, the paper discusses different pathlength calibration models and also investigates into the performance promotion of subtracted blood volume spectrometry. The details and main conclusions of the researches are as follows:1) The research applies tissue optics and chemometrics method to the analysis of different pathlength spectra equivalence. Firstly, the principle of spectral subtraction approach with different blood volume is derived. The origin of different pathlength blood spectra and the meaning ofacquire equivalentspectrum are analyzed. Then the spectral equivalence of different pathlength in non-scattering media is studied, and the normalized equivalent spectrum can be obtained using chemometric methods. After that, by analyzing the optical transmissionmodel in tissueoptics and the modifiedLambert-Beer’s law,the blood spectra with different pathlength is also equivalent. Finally, to prove the conclusion, experiments on glucose solution and intralipid were performed and equivalent spectra were obtained for each solution.2) To solve the problems occurred during the establishment of calibration models, a approach is proposed to improve the function of multivariate calibration model, which combines both orthogonal signal correction andnet analyte signal method. In this paper, analog spectrum is established with molar absorption coefficients of some biomolecules. The analyses of the analog spectra as well as various simulated solutions of glucose reveal that pathlength interference is restrained and the capability of multivariate calibration model is promoted.3) Analyze the practical performance of subtracted blood volume spectrometry through simulation solution experiment. To evaluate the effect of spectra subtraction, indicators such as mean Euclidean distance value and root-mean-square error are applied. To suppress the noise, algorithms of Savitzky-Galay smoothing, wavelet transform filtering and empirical mode decomposition filtering are adopted. Through these processes, the effect of spectra subtraction gets promoted. It is proves by related experiments that wavelet transform filtering and empirical mode decomposition filtering have an apparent advantage in the performance promotion.The researches of this paper provide both theoretical and experimental basis for the application of subtracted blood volume spectrometry in noninvasive biochemical sensing field.更多还原