Adaptive ensemble manifold learning for neuroimaging retrieval

In this paper, we proposed an adaptive ensemble manifold learning (AEML) for the neuroimaging retrieval via multimodality data. Specifically, we investigated the complementarity power of the multimodality data to index brain image disorder with different weights. Unlike the previous manifold learning methods, we develop a new learning method to incorporate the geometrical information and impose the similarity constraints. In addition, a generalized norm is also introduced to control the sparseness degree adaptively. Finally, a novel objective function is devised to retrieve the similar brain disorder in a database by a multi-task learning framework. We evaluated the proposed AEML algorithm via the public Alzheimer's Disease Neuroimaging Initiative (ADNI) database with 805 pre-diagnosed subjects. Experimental results show our algorithm achieved promising results and outperformed several related algorithms.