Ye Dong Hye, Litt Harold, Davatzikos Christos, Pohl Kilian M
SBIA, University of Pennsylvania, Philadelphia, USA.
Cardiovascular Imaging Section, University of Pennsylvania, Philadelphia, USA.
Funct Imaging Model Heart. 2011 May;6666:180-187. doi: 10.1007/978-3-642-21028-0_23.
This paper presents an image-based classification method, and applies it to classification of cardiac MRI scans of individuals with Tetralogy of Fallot (TOF). Clinicians frequently diagnose cardiac disease by measuring the ventricular volumes from cardiac MRI scans. Interrater variability is a common issue with these measurements. We address this issue by proposing a fully automatic approach for detecting structural changes in the heart. We first extract morphological features of each subject by registering cardiac MRI scans to a template. We then reduce the size of the features via a nonlinear manifold learning technique. These low dimensional features are then fed into nonlinear support vector machine classifier identifying if the subject of the scan is effected by the disease. We apply our approach to MRI scans of 12 normal controls and 22 TOF patients. Experimental result demonstrates that the method can correctly determine whether subject is normal control or TOF with 91% accuracy.
本文提出了一种基于图像的分类方法，并将其应用于法洛四联症（TOF）患者心脏磁共振成像（MRI）扫描的分类。临床医生经常通过测量心脏MRI扫描中的心室容积来诊断心脏病。这些测量中，评分者间的变异性是一个常见问题。我们通过提出一种用于检测心脏结构变化的全自动方法来解决这个问题。我们首先通过将心脏MRI扫描配准到一个模板来提取每个受试者的形态特征。然后，我们通过非线性流形学习技术减小特征的大小。接着，将这些低维特征输入到非线性支持向量机分类器中，以确定扫描的受试者是否患有该疾病。我们将我们的方法应用于12名正常对照者和22名TOF患者的MRI扫描。实验结果表明，该方法能够以91%的准确率正确判断受试者是正常对照还是TOF患者。
