Group-regularized individual prediction: theory and application to pain

摘要: Multivariate pattern analysis (MVPA) has become an important tool for identifying brain representations of psychological processes and clinical outcomes using fMRI related methods. Such methods can be used to predict or ‘decode’ states in individual subjects. Single-subject MVPA approaches, however, are limited by the amount quality individual-subject data. In spite higher spatial resolution, predictive accuracy from single-subject data often does not exceed what accomplished coarser, group-level maps, because patterns trained on amounts often-noisy Here, we present a method that combines population-level priors, form biomarker developed prior samples, with maps improve prediction. Theoretical results simulations motivate weighting based relative variances biomarker-based prediction—based groups—and individual-subject, cross-validated Empirical predicting pain activity trial-by-trial basis (single-trial prediction) across 6 studies (N = 180 participants) confirm theoretical predictions. Regularization biomarker—in this case, Neurologic Pain Signature (NPS)—improved prediction compared idiographic individuals' alone. The regularization scheme propose, which term group-regularized (GRIP), applied broadly within-person MVPA-based We also show how GRIP evaluate provide benchmarks appropriateness like NPS given study.