Hessian regularization based semi-supervised dimensionality reduction for neuroimaging data of Alzheimer´s disease

The neuroimaging data based computer-aided diagnosis of Alzheimer´s disease (AD) has attracted much attention. However, neuroimaging data is not only small sample size, but also only limited labeled samples. Therefore, the semi-supervised learning (SSL) has been applied for it. Recently, the Hessian regularization (HR) has been successfully applied to SSL. A newly proposed l_2,1 regularized correntropy algorithm for robust feature selection (CRFS) has achieved good performance for noise corrupted data, which is suitable for reducing the dimensions of neuroimaging data. We proposed a HR-based Semi-Supervised CRFS (HR-SSCRFS) algorithm, and then applied it to reduce the feature dimensions of neuroimaging data for classification of AD. The HR-SSCRFS was compared with LR-SSCRFS, supervised CRFS, and principal component analysis. The experimental results indicate that the proposed HR-SSCRFS significantly outperforms all other algorithms.