阿尔茨海默病磁共振图像的特征选择方法研究

发布时间：2018-04-14 19:04

本文选题：阿尔茨海默病 + 轻度认知障碍　；参考：《山东师范大学》2017年硕士论文

【摘要】：阿尔茨海默病(Alzheimer’s disease,AD),是一种最常见的神经系统退行性疾病。随着人口老龄化进程的加剧,阿尔茨海默病患者逐年大量增加,造成了十分巨大的社会影响。然而由于阿尔茨海默病的病变原因比较复杂,而且科学研究和临床检查没有足够的特异性,使得对该病早期的诊断准确率偏低。阿尔茨海默病的临床检查发展过程比较缓慢,能够有效进行早期无创敏感的诊断方法并不多。随着计算机图像处理方法和磁共振成像技术的发展,使用磁共振成像技术进行早期无创的体外检测颅内结构发展为一种有效可靠的方法。在临床上,磁共振成像对阿尔茨海默病的诊断和治疗有着非常重要的意义。本文利用磁共振成像数据,对阿尔茨海默病患病人群、轻度认知障碍人群以及正常老年人的大脑皮层的形态学进行了比较深入的研究。本论文研究内容和成果主要包括以下几个方面:首先,本文通过研究分析相关特征选择算法,得出一种新的特征选择算法,通过排名融合规则,将m RMR和Relief相结合,充分利用考虑Filter类特征选择方法的较高的计算效率以及利用Wrapper类特征选择方法使用分类效果作为评价标准具有较好的分类精度,使用新的特征选择方法对经过图像预处理的正常老年人和轻度认知障碍人群的磁共振数据进行特征选择,使得新的特征选择算法选出的特征子集具有较高的分类准确率。其次,为了使分类效果更优以及降低对融合过程的权重选择的时间复杂度,提出使用粒子群算法进行权重寻优,可以更快得到更好的权重系数,使得新的特征选择算法系统具有较低的时间复杂度。选出的特征对分类器具有更好的效果,并对算法进行了横向比较,在常用相关特征选择算法中具有比较明显的优势。最后,使用新的特征选择算法结合支持向量机进行特征选择及特征分类研究,选出具有最佳分类效果的特征并对比脑区图像,验证选出的特征对应的脑区部位主要与海马旁回相关,与前人的相关研究有一定相似性,同时也发现新的不相同的脑区左侧枕上和右侧楔前叶,为将来的研究提供了一定参考。
[Abstract]:Alzheimer's disease (AD) is one of the most common neurodegenerative diseases.With the aggravation of population aging, Alzheimer's disease patients are increasing year by year, resulting in great social impact.However, the early diagnostic accuracy of Alzheimer's disease is low due to the complexity of the disease and the lack of specificity in scientific research and clinical examination.The clinical examination of Alzheimer's disease develops slowly and there are few effective methods for early noninvasive and sensitive diagnosis.With the development of computer image processing and magnetic resonance imaging, it is an effective and reliable method to use magnetic resonance imaging to detect intracranial structures in vitro.Magnetic resonance imaging (MRI) plays an important role in the diagnosis and treatment of Alzheimer's disease.The morphology of cerebral cortex in patients with Alzheimer's disease, mild cognitive impairment and normal elderly was studied by magnetic resonance imaging (MRI).The main contents and achievements of this paper include the following aspects: firstly, by analyzing the relevant feature selection algorithms, a new feature selection algorithm is proposed, which combines m RMR and Relief by ranking fusion rules.Taking full advantage of the high computational efficiency of considering the Filter class feature selection method and using the Wrapper class feature selection method to use the classification effect as the evaluation criterion, it has good classification accuracy.The new feature selection method is used to select the magnetic resonance (MRI) data of the normal elderly and mild cognitive impairment population after image preprocessing, which makes the feature subset selected by the new feature selection algorithm have a higher classification accuracy.Secondly, in order to make the classification effect better and reduce the time complexity of weight selection in the fusion process, a particle swarm optimization algorithm is proposed for weight optimization, which can get better weight coefficients faster.The new feature selection algorithm system has lower time complexity.The selected features have a better effect on the classifier, and the algorithm is compared horizontally, which has obvious advantages in the common related feature selection algorithms.Finally, a new feature selection algorithm combined with support vector machine (SVM) is used to study the feature selection and feature classification. The features with the best classification effect are selected and compared with the brain image.The selected regions were mainly related to the parahippocampal gyrus and were similar to previous studies. At the same time, some new different regions of the brain were found in the left superior occipital area and the right anterior cuneate lobe, which provided a certain reference for future research.
【学位授予单位】：山东师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R749.16;TP391.41

【参考文献】