Feature ranking based nested support vector machine ensemble for medical image classification

• Published in: 2012 9th IEEE International Symposium on Biomedical Imaging (ISBI) 2012-01, p.146-149
• Main Authors: Varol, E., Gaonkar, B., Erus, G., Schultz, R., Davatzikos, C.
• Format: Article
• Language: English
• Subjects: Accuracy; Biomedical imaging; Classification; Diseases; Ensemble SVM; Feature extraction; Feature ranking; MRI; Support vector machines; Training; Variable speed drives
• ISSN: 1945-7928; 1945-8452
• DOI: 10.1109/ISBI.2012.6235505

This paper presents a method for classification of structural magnetic resonance images (MRI) of the brain. An ensemble of linear support vector machine classifiers (SVMs) is used for classifying a subject as either patient or normal control. Image voxels are first ranked based on the voxel wise t-statistics between the voxel intensity values and class labels. Then voxel subsets are selected based on the rank value using a forward feature selection scheme. Finally, an SVM classifier is trained on each subset of image voxels. The class label of a test subject is calculated by combining individual decisions of the SVM classifiers using a voting mechanism. The method is applied for classifying patients with neurological diseases such as Alzheimer's disease (AD) and autism spectrum disorder (ASD). The results on both datasets demonstrate superior performance as compared to two state of the art methods for medical image classification.