近红外光谱组合集成化学计量学模型的肠癌诊断方法研究

In a sense, the incidence and mortality rates of colorectal cancers are relatively high, to some extent, which is due to the lack of a diagnostic method with high sensitivity and specificity. However, both the sensitivity and specificity of a diagnostic method are key indicators, which maybe decide the health and life quality of tens of thousands of people. Thus, establishing a simple, rapid and accurate detection method for diagnosing and screening early colorectal cancer has important practical significance. This project mainly focuses at analyzing the characteristics of near-infrared spectra related to human colorectal cancer tissue, benign tissue and normal tissue, and clarifying the relationship between the near-infrared spectroscopic characteristics and bimolecular groups of various types of organizations. Further, in the frame of ensemble that proves to be a very effective field in our recent researches, based on chemometric methods such as multi-class support vector machines, it is expected to construct the diagnosis model with an output probability, accompanied by a spectrum database of cases and a basic frame of diagnosing colorectal cancers. Moreover, the strategies of sample set partitioning, the effect of training set size and the compromise between sensitivity and specificity will be investigated, so that the near-infrared spectroscopy technology can serve as a new method for automatic detection and early diagnosis of colorectal tumors.

大肠癌对人类健康威胁巨大，发病率和死亡率均较高，在某种程度上是由于目前仍缺乏一种具有高灵敏度和特异性的诊断方法，而灵敏度和特异性是诊断方法的关键指标，每提高一个百分点将有可能左右数万人的健康和生存质量，建立简单、快速和准确的检测方法对大肠癌的筛查和早期诊断具有重要的现实意义。本项目拟采用近红外光谱技术对人体大肠癌变组织、良性组织、正常组织进行光谱分析，探索各类组织的近红外分子光谱特征及其与生物分子基团的关联；并在我们近年来研究表明十分有效的集成框架下，辅以多分类支持向量机等化学计量学方法，力求建立初步的概率诊断模型、病例光谱数据库和大肠肿瘤光谱诊断基本框架。项目还将探索样本集划分、训练集规模影响、灵敏度和特异性折中处理等优化策略，有望推动近红外光谱技术成为大肠肿瘤自动检测和早期诊断的新方法。

癌症是一种严重危害人类健康乃至生命的疾病，它的发生是一个多因素、多阶段、复杂渐进的过程。肠癌在肿瘤的发病率中占第四，死亡率也很高，目前仍缺乏一种具有高灵敏度和特异性的诊断肠癌的方法，建立简单、快速和准确的检测方法对大肠癌的筛查和诊断具有重要的现实意义。本项目围绕肠癌诊断方法探索这一主题，收集和制备了大量人体肠组织样本，采用近红外光谱技术采集了几种模式的近红外光学信号，建立了一个初步的近红外光谱数据库。基于概率模型，小波变换，支持向量机，集成策略等新型化学计量学算法与模型，深入研究了大肠组织的近红外光谱特征、样本集划分、信号处理、信息提取及诊断模型构建与验证，该项目研究工作有助于推动近红外光谱技术成为肠癌自动检测和诊断的新方法。