基于三元泛函主成分分析和聚类分析的MRI图像分类研究

Research on Algorithms Based on Three Dimensional Functional Principal Component Analysis and Cluster Analysis for MRI Image Classification

一元泛函主成分分析(FPCA)已经在fMRI上成功进行了应用,但是目前很少有研究运用多元FPCA对MRI进行探索.本研究将一元FPCA推广到三元并应用于MRI的图像特征提取,并对提取的特征进行了后续研究,提出了一整套MRI病理及正常图像的分析方法.该方法的主要流程是先对MRI图像进行预处理(图像配准和图像分割),得到脑脊液图像,然后运用三元FPCA对脑脊液进行特征提取,再对提取的特征进行选择,随后利用k-means聚类算法对特征进行聚类,来判断图像所属的类别(正常或异常),从而达到颅脑MRI图像病变筛查的目的.将该方法应用于颅脑MRI快速自旋回波T2加权像中,结果发现,相比于传统PCA,三元FPCA展现出更好的特征提取能力,可以有效提高图像分类的准确率.

One dimensional functional principal component analysis（FPCA）has been successfully applied in functional magnetic resonance imaging（fMRI）data.But there is scarce study focused on the application of high dimensional FPCA in magnetic resonance imaging（MRI）data.Three dimensional FPCA extended from one dimensional FPCA was provided and its application in MRI data to extract features was studied.A full set of techniques used to discriminate between pathological images and normal images was designed.It mainly consisted of the following three steps.Firstly,MRI data were preprocessed.The preprocessing steps included image registration and segmentation.The cerebrospinal fluid was obtained after image preprocessing.Secondly,functional PCA was employed to extract features from the segmented images.After that a feature selection method was performed for the extracted features.Thirdly,binary classification results were derived by k-means clustering method using features selected,which reached the goal to screen diseases.The application of this new method to brain MR turbo spin echo T2 weighted image data showed that FPCA outperformed multivariate PCA in feature extraction and classification accuracy.