ICDM '05 Proceedings of the Fifth IEEE International Conference on Data Mining
Feature combination using boosting
Pattern Recognition Letters
LIBSVM: A library for support vector machines
ACM Transactions on Intelligent Systems and Technology (TIST)
Hi-index | 0.00 |
FDG-PET imaging offers the potential for an image-based automated identification of different dementia syndromes. However, various global and local FDG-PET image features have their limitations in characterizing the patterns of this disease. In this paper, we propose an automated approach to identifying the patients with suspected Alzheimer's disease, patients with frontotemporal dementia and normal controls based on the jointly using a group of global features and three groups of local features extracted from parametric FDG-PET images. In this approach, we employ the genetic algorithm to select the features that have best discriminatory ability, and use the AdaBoost technique to adaptively combine four feature groups in constructing a strong classifier. We compared our approach to other classification methods in 154 clinical FDG-PET studies. Our results show that, with the complementary use of the selected global and local features, the proposed approach can substantially improve the accuracy of FDG-PET imaging-based dementia identification.