Title: Supervised sparse hierarchical components analysis with application to resting-state functional MRI data
Authors: Atsushi Kawaguchi - Saga University (Japan) [presenting]
Abstract: Brain functional connectivity is useful for identifying biomarkers that can be used for diagnosis of brain disorders. This connectivity can be measured using resting-state functional Magnetic Resonance Imaging (rs-fMRI). Previous studies were based on a sequential application of the graphical model for network estimation and machine learning for constructing a prediction formula for the outcome (e.g., disease or healthy) from the estimated network. This approach results in a less informative network for diagnosis because the network is estimated independently of the outcome. A regression method with the score from the rs-fMRI based on the supervised sparse hierarchical components analysis (SSHCA) is proposed. SSHCA has a hierarchical structure consisting of the network model (block scores at the individual level) and the scoring model (super scores at the population level). The multiple logistic regression model with super scores as the predictor was used to estimate the diagnostic probability. An advantage of the proposed method is that the outcome-related (supervised) network connection and the multiple scores corresponding to sub-network estimation will be helpful in interpretation. The proposed method was applied to the data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) and was compared with the existing method based on the sequential approach.