Group-regularized individual prediction

Theory and application to pain

Martin A. Lindquist, Anjali Krishnan, Marina López-Solà, Marieke Jepma, Choong Wan Woo, Leonie Koban, Mathieu Roy, Lauren Y. Atlas, Liane Schmidt, Luke J. Chang, Elizabeth A. Reynolds Losin, Hedwig Eisenbarth, Yoni K. Ashar, Elizabeth Delk, Elizabeth Losin

Research output: Contribution to journalArticle

19 Citations (Scopus)

Abstract

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

Original languageEnglish (US)
JournalNeuroImage
DOIs
StateAccepted/In press - 2016

Fingerprint

Multivariate Analysis
Pain
Biomarkers
Nervous System
Population
Psychology
Benchmarking
Brain
Magnetic Resonance Imaging

Keywords

  • Empirical Bayes
  • FMRI
  • Machine learning
  • Mega-analysis
  • Meta-analysis
  • MVPA
  • Pain
  • Prediction
  • Shrinkage
  • Statistical learning

ASJC Scopus subject areas

  • Cognitive Neuroscience
  • Neurology

Cite this

Lindquist, M. A., Krishnan, A., López-Solà, M., Jepma, M., Woo, C. W., Koban, L., ... Losin, E. (Accepted/In press). Group-regularized individual prediction: Theory and application to pain. NeuroImage. https://doi.org/10.1016/j.neuroimage.2015.10.074

Group-regularized individual prediction : Theory and application to pain. / Lindquist, Martin A.; Krishnan, Anjali; López-Solà, Marina; Jepma, Marieke; Woo, Choong Wan; Koban, Leonie; Roy, Mathieu; Atlas, Lauren Y.; Schmidt, Liane; Chang, Luke J.; Reynolds Losin, Elizabeth A.; Eisenbarth, Hedwig; Ashar, Yoni K.; Delk, Elizabeth; Losin, Elizabeth.

In: NeuroImage, 2016.

Research output: Contribution to journalArticle

Lindquist, MA, Krishnan, A, López-Solà, M, Jepma, M, Woo, CW, Koban, L, Roy, M, Atlas, LY, Schmidt, L, Chang, LJ, Reynolds Losin, EA, Eisenbarth, H, Ashar, YK, Delk, E & Losin, E 2016, 'Group-regularized individual prediction: Theory and application to pain', NeuroImage. https://doi.org/10.1016/j.neuroimage.2015.10.074
Lindquist, Martin A. ; Krishnan, Anjali ; López-Solà, Marina ; Jepma, Marieke ; Woo, Choong Wan ; Koban, Leonie ; Roy, Mathieu ; Atlas, Lauren Y. ; Schmidt, Liane ; Chang, Luke J. ; Reynolds Losin, Elizabeth A. ; Eisenbarth, Hedwig ; Ashar, Yoni K. ; Delk, Elizabeth ; Losin, Elizabeth. / Group-regularized individual prediction : Theory and application to pain. In: NeuroImage. 2016.
@article{c6d8b6722c664c07b03892ab72b8197b,
title = "Group-regularized individual prediction: Theory and application to pain",
abstract = "Multivariate pattern analysis (MVPA) has become an important tool for identifying brain representations of psychological processes and clinical outcomes using fMRI and related methods. Such methods can be used to predict or 'decode' psychological states in individual subjects. Single-subject MVPA approaches, however, are limited by the amount and quality of individual-subject data. In spite of higher spatial resolution, predictive accuracy from single-subject data often does not exceed what can be accomplished using coarser, group-level maps, because single-subject patterns are trained on limited amounts of often-noisy data. Here, we present a method that combines population-level priors, in the form of biomarker patterns developed on prior samples, with single-subject MVPA maps to improve single-subject prediction. Theoretical results and simulations motivate a weighting based on the relative variances of biomarker-based prediction-based on population-level predictive maps from prior groups-and individual-subject, cross-validated prediction. Empirical results predicting pain using brain activity on a trial-by-trial basis (single-trial prediction) across 6 studies (N = 180 participants) confirm the theoretical predictions. Regularization based on a population-level biomarker-in this case, the Neurologic Pain Signature (NPS)-improved single-subject prediction accuracy compared with idiographic maps based on the individuals' data alone. The regularization scheme that we propose, which we term group-regularized individual prediction (GRIP), can be applied broadly to within-person MVPA-based prediction. We also show how GRIP can be used to evaluate data quality and provide benchmarks for the appropriateness of population-level maps like the NPS for a given individual or study.",
keywords = "Empirical Bayes, FMRI, Machine learning, Mega-analysis, Meta-analysis, MVPA, Pain, Prediction, Shrinkage, Statistical learning",
author = "Lindquist, {Martin A.} and Anjali Krishnan and Marina L{\'o}pez-Sol{\`a} and Marieke Jepma and Woo, {Choong Wan} and Leonie Koban and Mathieu Roy and Atlas, {Lauren Y.} and Liane Schmidt and Chang, {Luke J.} and {Reynolds Losin}, {Elizabeth A.} and Hedwig Eisenbarth and Ashar, {Yoni K.} and Elizabeth Delk and Elizabeth Losin",
year = "2016",
doi = "10.1016/j.neuroimage.2015.10.074",
language = "English (US)",
journal = "NeuroImage",
issn = "1053-8119",
publisher = "Academic Press Inc.",

}

TY - JOUR

T1 - Group-regularized individual prediction

T2 - Theory and application to pain

AU - Lindquist, Martin A.

AU - Krishnan, Anjali

AU - López-Solà, Marina

AU - Jepma, Marieke

AU - Woo, Choong Wan

AU - Koban, Leonie

AU - Roy, Mathieu

AU - Atlas, Lauren Y.

AU - Schmidt, Liane

AU - Chang, Luke J.

AU - Reynolds Losin, Elizabeth A.

AU - Eisenbarth, Hedwig

AU - Ashar, Yoni K.

AU - Delk, Elizabeth

AU - Losin, Elizabeth

PY - 2016

Y1 - 2016

N2 - Multivariate pattern analysis (MVPA) has become an important tool for identifying brain representations of psychological processes and clinical outcomes using fMRI and related methods. Such methods can be used to predict or 'decode' psychological states in individual subjects. Single-subject MVPA approaches, however, are limited by the amount and quality of individual-subject data. In spite of higher spatial resolution, predictive accuracy from single-subject data often does not exceed what can be accomplished using coarser, group-level maps, because single-subject patterns are trained on limited amounts of often-noisy data. Here, we present a method that combines population-level priors, in the form of biomarker patterns developed on prior samples, with single-subject MVPA maps to improve single-subject prediction. Theoretical results and simulations motivate a weighting based on the relative variances of biomarker-based prediction-based on population-level predictive maps from prior groups-and individual-subject, cross-validated prediction. Empirical results predicting pain using brain activity on a trial-by-trial basis (single-trial prediction) across 6 studies (N = 180 participants) confirm the theoretical predictions. Regularization based on a population-level biomarker-in this case, the Neurologic Pain Signature (NPS)-improved single-subject prediction accuracy compared with idiographic maps based on the individuals' data alone. The regularization scheme that we propose, which we term group-regularized individual prediction (GRIP), can be applied broadly to within-person MVPA-based prediction. We also show how GRIP can be used to evaluate data quality and provide benchmarks for the appropriateness of population-level maps like the NPS for a given individual or study.

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

KW - Empirical Bayes

KW - FMRI

KW - Machine learning

KW - Mega-analysis

KW - Meta-analysis

KW - MVPA

KW - Pain

KW - Prediction

KW - Shrinkage

KW - Statistical learning

UR - http://www.scopus.com/inward/record.url?scp=84955254159&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84955254159&partnerID=8YFLogxK

U2 - 10.1016/j.neuroimage.2015.10.074

DO - 10.1016/j.neuroimage.2015.10.074

M3 - Article

JO - NeuroImage

JF - NeuroImage

SN - 1053-8119

ER -