基于像素特征学习的磁共振图像中β淀粉样蛋白沉积信息检测算法

发布时间：2018-03-31 17:22

本文选题：阿尔茨海默症　切入点：β淀粉样蛋白沉积　出处：《生物医学工程学杂志》2017年03期

【摘要】：β淀粉样蛋白(Aβ)沉积是阿尔茨海默症(AD)的重要防治靶点,在脑中及早发现Aβ蛋白沉积是AD早期诊断的关键。磁共振成像(MRI)是一种理想成像方式,但不能直接显示图像中存在的沉积信息。本文基于过滤式和封装式的选择模式引入链式智能体遗传算法(CAGA)、主成分分析(PCA)、支持向量机(SVM)和随机森林(RF),构建6种特征学习分类算法,通过像素特征优选来实现Aβ蛋白沉积信息(分布)的检测。首先,分割脑磁共振(MR)图像中的脑组织;然后提取脑组织中的像素值形成像素特征向量;接着设计特征学习分类算法对像素实现特征优选,并基于投票机制得到一组最终最优特征向量;最后采用弹性映射方法将最优像素特征向量映射到脑MR图像上,并标记出对应的像素点,从而显示出Aβ蛋白沉积的分布。实验结果表明,本文的像素特征学习方法可提取并显示Aβ蛋白沉积信息,最高分类准确率可达到80%以上,表明该方法是可行和有效的。本文从脑MR图像中检测的Aβ沉积信息将有助于提高基于MR的AD诊断准确率。
[Abstract]:Amyloid A 尾 deposition is an important prevention and treatment target for Alzheimer's disease (AD). Early detection of A 尾 protein deposition in the brain is the key to early diagnosis of AD. Magnetic resonance imaging (MRI) is an ideal imaging method. In this paper, we introduce chain agent genetic algorithm, principal component analysis (PCA), support vector machine (SVMN) and random forest RFRV to construct six feature learning classification algorithms, which are based on filter and encapsulation selection mode, including chain agent genetic algorithm (CAGAA), principal component analysis (PCA), support vector machine (SVM), and random forest fission (RFRF). The detection of A 尾 -protein deposition information (distribution) is realized by pixel feature optimization. Firstly, the brain tissue is segmented from the brain MRI image, then the pixel value in the brain tissue is extracted to form the pixel feature vector. Then the feature learning classification algorithm is designed to select the pixels, and a set of final optimal feature vectors is obtained based on the voting mechanism. Finally, the optimal pixel feature vector is mapped to the brain Mr image by elastic mapping method. The results show that the pixel feature learning method in this paper can extract and display the A 尾 protein deposition information, and the highest classification accuracy can reach more than 80%. It is shown that this method is feasible and effective. The A 尾 deposition information detected from brain Mr images in this paper will help to improve the accuracy of AD diagnosis based on Mr.
【作者单位】：重庆大学通信工程学院;重庆大学脑科学协同创新中心;第三军医大学医学图像学教研室;
【基金】：国家自然科学基金资助项目(61108086,91438104,61571069,81601970,61501065) 西南医院联合孵化项目(SWHLHYS-11) 重庆市社会事业与民生保障科技创新专项(cstc2016shmszx40002) 重庆市基础与前沿研究项目(cstc2016jcyj A0043,cstc2016jcyj A0064,cstc2016jcyj A0134) 重庆市教委项目(KJ1603805) 中央高校基本科研业务费专项资金资助项目(10611CDJXZ238826)
【分类号】：R749.16;TP391.41

【相似文献】