基于功能磁共振成像的Ⅱ型糖尿病患者认知功能预测分析

发布时间：2018-06-11 11:53

本文选题：Ⅱ型糖尿病 + 特征提取　；参考：《郑州大学》2017年硕士论文

【摘要】：Ⅱ型糖尿病是一种常见的代谢疾病,患者通常伴有认知能力下降且老年痴呆的风险也大幅增加。目前,在医学上采用蒙特利尔认知评估量表(MoCA),对患者进行认知功能异常的评定,但这种方法易受教育程度、文化差异、检查者技巧和经验的影响。缺乏一种易操作、定量的对认知功能评定的方法,来避免以上问题带来干扰。功能磁共振成像(functional magnetic resonance imaging,fMRI)有现有的大脑成像技术相比,具有更高的时间和空间分辨率,从fMRI可以提取大量高维的影像特征,来研究大脑与记忆力和注意力的关系。目前,基于fMRI对Ⅱ型糖尿病患者认知功能预测的研究较少,因此,本文提出了一种新的解决思路,从fMRI中提取特征,利用机器学习方法建立认知功能预测模型。文中实验数据来源于河南省人民医院,其中包括95例fMRI影像及临床信息。实验的主要流程是,从影像中提取特征,使用弹性网(Elastic Net)方法对影像特征及临床特征进行特征降维,用支持向量机(Support Vector Machine,SVM)和逻辑斯特回归(Logistic Regression,LR)两种算法构建认知功能预测模型,在模型训练的过程中,采用十折交叉验证确定模型参数。支持向量机分类模型正确率达到93.24%,AUC为0.9781,召回率为92.11%;逻辑斯特回归分类模型正确率达到了95.95%,AUC为0.9971,召回率为94.74%。通过对比,发现逻辑斯特回归模型的预测效果更好。相对于传统Ⅱ型糖尿病患者认知障碍评定方法,通过从影像中提取特征建立预测模型,具有不受外界因素的干扰的优点,与此同时,也为认知功能异常的评定给出了一种新的方法,另外实验中通过弹性网模型降维后选择出来影像特征,可以对Ⅱ型糖尿病的临床诊断和治疗提供了重要线索。
[Abstract]:Type 2 diabetes is a common metabolic disease, with cognitive impairment and a significant increase in the risk of dementia. At present, the Montreal Cognitive Assessment scale (MOCAA) is used to evaluate the cognitive dysfunction of patients. However, this method is susceptible to the influence of educational level, cultural differences, examiner skills and experience. Lack of an easy-to-operate, quantitative assessment of cognitive function to avoid interference with these problems. Functional magnetic resonance imaging of MRI (functional magnetic resonance imaging) has higher temporal and spatial resolution than existing brain imaging techniques. It can extract a large number of high-dimensional image features from fMRI to study the relationship between brain and memory and attention. At present, there is little research on the prediction of cognitive function based on fMRI in patients with type II diabetes. Therefore, a new approach is proposed to extract features from fMRI and establish a cognitive function prediction model using machine learning method. The experimental data were obtained from Henan Provincial people's Hospital, including 95 fMRI images and clinical information. The main flow of the experiment is to extract the feature from the image, to reduce the dimension of the feature and clinical feature by using the elastic net method, and to construct the cognitive function prediction model by using the support vector machine support Vector Machine (SVM) and the logistic regression model (LRR). In the process of model training, ten fold cross validation is used to determine the model parameters. The correct rate of classification model of support vector machine is 93.24 and AUC is 0.978 1 and recall rate is 92.11. The correct rate of classification model of logistic regression is 95.95 AUC of 0.9971 and recall rate of 94.74. By comparison, it is found that the prediction effect of the logistic regression model is better. Compared with the traditional assessment method of cognitive impairment in type 2 diabetes mellitus patients, the prediction model by extracting features from the image has the advantage of not being interfered by external factors, at the same time, It also provides a new method for the evaluation of cognitive dysfunction. In addition, it can provide an important clue for the clinical diagnosis and treatment of type 2 diabetes mellitus by reducing the dimension of the elastic network model and selecting the image features in the experiment.
【学位授予单位】：郑州大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R587.2;R445.2;TP391.41

【相似文献】