基于数据挖掘的阿尔茨海默症风险因素的动态模式探测研究

项目名称： 基于数据挖掘的阿尔茨海默症风险因素的动态模式探测研究

项目编号： No.61300107

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 自动化技术、计算机技术

项目作者： 曾安

作者单位： 广东工业大学

项目金额： 22万元

中文摘要： 作为一种老年人常见的神经系统变性疾病，阿尔茨海默症(AD)因其高发病率和较差预后，正在给全社会带来越来越沉重的负担。鉴于AD病理的复杂性，目前尚无可靠的诊断方法对其进行早期临床诊断。当前AD流行病学研究主要集中在利用Logistic回归和Cox回归进行建模以期揭示出多风险因素与AD的关联关系。但这些预测模型很难同时展示出高的灵敏度和特异性。为提高预测效果，本课题首先设计出基于粗集理论和聚类技术的启发性知识获取算法；接着提出基于数据挖掘技术的数据修补集成方法、基于启发性知识的评分函数、组合极值优化方法以及基于此的搜索策略和粒子退化处理技术、基于"分而治之"思想的数据划分方法，设计出一套针对本项目数据特点（数据量大、数据缺失和混合变量等）的动态贝叶斯网学习和推理算法以探测出各风险因素与AD患病风险之间的动态非线性关联模式；最后结合临床应用构建出适合于我国AD高危人群筛查与临床早期诊断辅助系统。

中文关键词： 阿尔茨海默症；贝叶斯网络；风险因素；深度学习；衰弱模型

英文摘要： Alzheimer's Disease (AD) is the most common type of dementia in late life and has become a heavy burden for society and the economy. So far, a definite diagnosis of AD has to rely on autopsy, while a comprehensive clinical assessment is typically performed to diagnose a probable or possible AD. Currently, research on AD diagnosis and prognosis has mainly focused on the use of Logistic regression and Cox regression analyses in building models to demonstrate the relationship between multiple risk factors and AD incidence. Using these methods, it has been difficult to produce a predictive model that can achieve both high sensitivity and high specificity simultaneously. To further improve the prediction performance, this project will design a heuristic knowledge acquisition algorithm based on rough set theory and the clustering technique; and then, propose the missing-data-handling ensemble method based on data mining techniques, the scoring functions based on heuristic knowledge, the optimization method based on combinations of extreme elements, search strategy and particle degeneration processing technique based on the optimization method, data partition method based on divide-and-conquer; finally, advocate learning and reasoning algorithms for Dynamic Bayesian Networks which adapt to a large volume of real-life c

英文关键词： Alzheimer's Disease；Bayesian network；Risk factor；deep learning；Frailty model