A Graph-Based Decoding Model for Incomplete Multi-Subject fMRI Functional Alignment
share
As a successful application of multi-view learning, Hyperalignment and Shared Response Model are two effective functional alignment methods of neural activities across multiple subjects. Though they have been studied widely and can significantly improve functional Magnetic Resonance Imaging (fMRI) analysis, they are not able to tackle various kinds of fMRI datasets today, especially when they are incomplete, i.e., some of the subjects probably lack the responses to some stimuli or different subjects might follow different sequences of stimuli. In this paper, a cross-view graph that assesses the connection between any two samples across subjects is taken as an anchor for developing a more flexible framework that suits an assortment of fMRI datasets. To handle large-scale datasets, a kernelbased optimization that allows for non-linear feature extraction is theoretically developed for the proposed framework. Further, the proposed optimization allows us to do Principal Component Analysis, which can filter a specific Gaussian noise, in the new feature space with any kernel. Empirical studies confirm that the proposed method under both incompleteness and completeness can achieve better performance than other state-of-the-art functional alignment methods without incompleteness
READ FULL TEXT