Hirotaka Yamagata

This study compares unsupervised feature selection with supervised methods to identify robust resting-state functional connectivity patterns associated with psychiatric disorders.

The paper evaluates rs-fMRI analysis pipelines for MDD biomarkers using multi-site datasets, finding optimal combinations of parcellation, connectivity estimation, harmonization, and machine learning methods.

The paper identifies common brain networks between major depressive disorder and depression symptoms that generalize across independent imaging sites.

The paper develops a harmonization method for resting-state fMRI data across sites by separating site differences into sampling and measurement biases, reducing measurement bias by 29% and improving signal-to-noise ratios by 40%.