On the reproducibility of in vivo temporal signal-to-noise ratio and its utility as a predictor of subject-level t-values in a functional magnetic resonance imaging study.