Wrapper-based Feature Learning Method used for Alzheimer's disease Identification

In this paper, a wrapper-based feature learning method is presented for Alzheimer's disease (AD) identification. First, with the extracted features, the feature importance of AD and Healthy Controls (HC) is ranked from high to low according to sensitivity and specificity of classification. Second, the AD and HC feature subsets are formed using forward incremental feature selection method. Finally, the weighted classification method is used to weight the probability estimates of AD and HC to obtain the final classification performance. Experimental results show that the wrapper-based feature learning method can select most important feature parameters or featured region of interests (ROIs) of brain compared with other feature selection methods. The proposed method will be useful in finding featured ROIs of brain that most related to AD identification, which can further improve the AD identification performance.