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

Detection algorithm of amyloid β-protein deposition in magnetic resonance image based on pixel feature learning method

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。

Amyloid β-protein （Aβ） deposition is an important prevention and treatment target for Alzheimer’s disease （AD）, and early detection of Aβ deposition in the brain is the key to early diagnosis of AD. Magnetic resonance imaging （MRI） is the perfect imaging technology for the clinical diagnosis of AD, but it cannot display the plaque deposition directly. In this paper, based on two feature selection modes-filter and wrapper, chain-like agent genetic algorithm （CAGA）, principal component analysis （PCA）, support vector machine （SVM） and random forest （RF）, we designed six kinds of feature learning classification algorithms to detect the information （distribution） of Aβ deposition through magnetic resonance image pixels selection. Firstly, we segmented the brain region from brain MR images. Secondly, we extracted the pixels in the segmented brain region as a feature vector （features） according to rows. Thirdly, we conducted feature learning on the extracted features, and obtained the final optimal feature subset by voting mechanism. Finally, using the final optimal selected features, we could find and mark the corresponding pixels on the MR images to show the information about Aβ plaque deposition by elastic mapping. According to the experimental results, the proposed pixel features learning methods in this paper could extract and reflect Aβ plaque deposition, and the best classification accuracy could be as high as 80%, thereby showing the effectiveness of the methods. The proposed methods can precisely detect the information of the Aβ plaque deposition, thereby being helpful for improving classification accuracy of diagnosis of AD.