基于蛋白组学和MRI脑图像纹理的早期阿尔茨海默症预测模型研究

发布时间：2018-04-30 08:23

本文选题：阿尔兹海默 + 血浆蛋白　；参考：《首都医科大学》2017年硕士论文

【摘要】：背景阿尔兹海默(Alzheimer’s Disease,AD)是一种慢性的神经系统退行性疾病,该疾病的临床表现主要为为记忆方面的退化以及认知功能障碍,严重威胁老年人的生活质量和生命安全。欧美国家中65岁以上老年人的痴呆症患病率约为4~8%,我国的痴呆患病率为7.8%,其中AD的患病率为4.8%。作为世界上人口数量最为庞大的国家,我国将面临愈发严峻的人口老龄化问题,AD将给患者,家庭和社会带来极大沉重的经济和生活负担。AD从出现临床症状至首次确诊时间平均大于一年,且病情多为中重度(67%),早期AD诊断的研究一直是国内外的热点和难点问题之一。AD发展涉及脑部结构微小且复杂的变化,通过图像微观纹理特征预测AD进展具有潜在价值。磁共振成像(Magnetic Resonance Imaging,MRI)的使用已被证明在预测轻度认知功能损害(Mild Cognitive Impairment,MCI)到AD的转换以及老年人认知功能下降方面上有很大的提示作用。同样,血浆蛋白组学已经被证明具有诊断AD以及预测MCI转换为AD的珍贵价值。结合血浆蛋白质组学和MRI成像作为生物标志物在早期AD诊断与预测中具有潜在优势。在MRI图像用于临床疾病识别或预测的模型研究中,高斯过程(Gaussian Processes,GP)分类表现出强大的能力。高斯过程是基于统计学习理论和贝叶斯理论发展起来的一种有监督的机器学习算法,高斯过程泛化能力强,超参数设置灵活、具有非参数推断以及概率输出等优点,适用于处理非线性和高维度等复杂回归问题。目的通过基于Contourlet变换提取脑部图像微观纹理特征,结合血浆蛋白组学生物标志物,采用高斯过程建立阿尔茨海默症的早期预测模型,为AD的早期诊断提供相关证据。方法本次研究共收集420例数据,其中AD患者84例,MCI患者287例,正常对照组49例。采用区域增长法从冠状位的脑部MRI图像中分割得到海马区域,采用Contourlet变换处理对海马区域图像进行处理并且计算14个纹理值参数。基于基线数据,学采用t检验或方差分析进行组间差异性比较,对血浆蛋白组间差异有统计学意义的变量采用LASSO(Least Absolute Shrinkage and Selection Operator)回归进行变量筛选,然后采用高斯过程模型以及支持向量机模型进行分类建模,考虑组合核函数选择最佳分类模型并做交叉验证;基于MCI患者基线基本信息,血浆蛋白数据和脑图像数据,以随访期内是否转换为AD作为结局标签进行建模,建立早期AD的分类预测模型。结果对于AD和健康对照组基线血浆蛋白浓度的t检验比较得到Apo AII,FSH,FASLG receptor等18种血浆蛋白组间差异有统计学意义;AD组以及健康对照组随访1年期前后,64种血浆蛋白有组间差异有统计学意义。LASSO回归分析得到20种血浆蛋白可作为早期AD诊断潜在的生物标志物,灵敏度76.2%,特异度81.3%。ROC曲线下面积为80.4%(95%CI:86.2%~79%);以MCI组是否转换为AD为结局,对有组间差异的血浆蛋白进行LASSO回归,并校正了性别和年龄,得到BNP,IL16,TBG,APOE,PLGF,TFF3等6种血浆蛋白,灵敏度91.2%,特异度78.4%,ROC曲线下面积为84.1%(95%CI:91.8%~81.6%)。结合研究对象的基本信息,分别基于左右测以及双侧海马脑图像纹理特征建立高斯过程分类模型。基于AD和健康对照两组构建的分类模型,右侧海马区的灵敏度为91.2%,特异度为81.6%,大于左侧海马区域模型;基于双侧海马的分ROC曲线下面积(0.922)大于基于左右侧海马区图像单独建立的模型(0.851和0.901)。基于MCI基线数据和随访结局建立的预测模型中,GR模型预测MCI转化的准确率达到88.4%,预测MCI转归为正常的准确率达到80.0%。SVM模型预测MCI转化的准确率达到81.0%,预测MCI转归为正常的准确率达到60.0%。结论血浆蛋白水平的IL-16,TBG,BNP,TFF3,PLGF和ApoE的组合可以区分AD患者和健康个体可以用于早期诊断和监测AD以及预测MCI转化为AD;具有高斯径向基核函数的组合核函数高斯过程方法预测效果较好;基于MR图像纹理以及血浆蛋白数据构建预测模型,对AD的早期预测具有积极作用。
[Abstract]:Background Alzheimer 's Disease (AD) is a chronic neurodegenerative disease. The clinical manifestations of this disease are mainly memory degradation and cognitive impairment, which seriously threaten the quality of life and life safety of the elderly. The prevalence rate of dementia in older people over 65 years in Europe and America is about 4~8%, The prevalence rate of dementia in the country is 7.8%, of which the prevalence of AD is 4.8%. as the largest population in the world. Our country will face the increasingly severe problem of population aging. AD will bring great heavy economic and living burden to patients, families and society, and the.AD from the appearance of bed symptoms to the first diagnosis is more than one year, and Most of the disease is moderate to severe (67%), the early AD diagnosis has been one of the hot and difficult problems at home and abroad..AD development involves small and complex changes in the brain structure. It is of potential value to predict the progress of AD through the microscopic texture features of the image. The use of Magnetic Resonance Imaging (MRI) has been proved to be a mild recognition in the prediction. The Mild Cognitive Impairment (MCI) has a great hint in the conversion of AD and the decline of cognitive function in the elderly. Similarly, plasma proteomics has been proved to have a valuable value in the diagnosis of AD and the prediction of MCI conversion to AD. Combined plasma proteomics and MRI imaging as a biomarker in early AD It has a potential advantage. In the study of MRI images for clinical disease identification or prediction, the Gauss process (Gaussian Processes, GP) classification shows strong ability. The Gauss process is a supervised machine learning algorithm based on statistical learning theory and Bias theory, and the generalization ability of Gauss process. Strong, super parameter setting is flexible, has the advantages of non parametric inference and probability output. It is suitable for dealing with complex regression problems such as nonlinear and high dimension. Objective to establish the early precondition of Alzheimer's disease by using the Gauss process to extract the microscopic texture features of brain images based on Contourlet transform and combine the biomarkers of plasma proteomics. The test model provided relevant evidence for the early diagnosis of AD. Methods a total of 420 data were collected in this study, including 84 cases of AD patients, 287 cases of MCI patients and 49 normal controls. The hippocampus region was segmented by regional growth method from the MRI image of the coronal position, and the hippocampal region images were processed by Contourlet transformation treatment and 1 were calculated. 4 texture value parameters. Based on the baseline data, t test or ANOVA were used to compare the differences between groups. The variables with significant differences in plasma protein groups were selected by LASSO (Least Absolute Shrinkage and Selection Operator) regression, and then the Gauss process model and support vector machine model were used. The optimal classification model of the combined kernel function was selected and the cross validation was taken into consideration. Based on the basic information of the MCI baseline, the plasma protein data and the brain image data, the model was modeled as the outcome label for the conversion of AD in the follow-up period, and the early AD classification prediction model was established. The results were for the baseline plasma eggs of the AD and the healthy control groups. The t test of white concentration showed that there were significant differences between the 18 plasma protein groups, such as Apo AII, FSH, FASLG receptor, and so on, and in the AD group and the healthy control group, before and after the 1 years of follow-up, the differences between the 64 plasma proteins were statistically significant and the 20 plasma proteins could be used as a potential biomarker for early AD diagnosis. Degree 76.2%, the area under the specificity 81.3%.ROC curve was 80.4% (95%CI:86.2%~79%), and if the group MCI was converted to AD, the plasma protein with different groups was returned by LASSO, and the sex and age were corrected, and 6 plasma proteins, such as BNP, IL16, TBG, APOE, PLGF, TFF3, were obtained. The sensitivity was 91.2%, the specificity was 78.4%, and the area under the ROC curve was 84.1%. 91.8%~81.6%). Based on the basic information of the subjects, the Gauss process classification model was established on the basis of the left and right tests and the texture features of the bilateral hippocampal images. The sensitivity of the right hippocampal region was 91.2% and the specificity was 81.6%, which was greater than the left hippocampal region model based on the two groups of AD and the healthy control groups. The sub R based on the bilateral hippocampus was divided into two groups. The area under the OC curve (0.922) is larger than the model based on the left and right hippocampal images (0.851 and 0.901). In the prediction model based on the MCI baseline data and the follow-up outcome, the GR model predicts the accuracy of MCI conversion to 88.4%, and the prediction of MCI to the normal accuracy reaches the 80.0%.SVM model to predict the MCI transformation accuracy of 81. %, the prediction of MCI to the normal accuracy rate of 60.0%. to the plasma protein level of IL-16, TBG, BNP, TFF3, PLGF and ApoE can distinguish between AD patients and healthy individuals can be used for early diagnosis and monitoring AD and predictive MCI into AD; Gauss radial basis kernel function of the combined kernel function Gauss process method prediction effect is better; The prediction model based on MR image texture and plasma protein data has a positive effect on the early prediction of AD.

【学位授予单位】：首都医科大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R749.16;R445.2

【相似文献】