Classification of Alzheimer's Disease Based on Cortical Thickness Using AdaBoost and Combination Feature Selection Method

In this research, using the idea of ensemble, we designed and applied a new supervised learning algorithm for classification of Alzheimer's disease (AD). Using MRI cortical surface-based analysis, cortical thickness of AD patients and normal controls were measured. All these data were retrieved from Alzheimer's Disease Neuroimaging Initiative (ADNI) database. We mainly used the cortical thickness data in our research. As human brains can be divided into many lobes and regions, which are of different distinctive capability, we adopted an ensemble feature selection method that filters these lobes according to their discriminative ability, and randomly selects the features for certain times to create several subsets. Each part of the data owns a classifier for training. And then we combined all the classifiers to form a more powerful classifier using AdaBoost. Linear discriminate analysis were used to build up these classifiers. The generalization accuracy using test data set can achieve about 0.86 if selected the parameters well. Our classification method based on ensemble feature selection was therefore proposed and could be used in AD classification problems or other related areas.