Bayesian multi-task learning for decoding multi-subject neuroimaging data

Andre F Marquand,Michael Brammer,Steven C.R Williams,Orla M Doyle

doi:10.1016/j.neuroimage.2014.02.008

Abstract

Decoding models based on pattern recognition (PR) are becoming increasingly important tools for neuroimaging data analysis. In contrast to alternative (mass-univariate) encoding approaches that use hierarchical models to capture inter-subject variability, inter-subject differences are not typically handled efficiently in PR. In this work, we propose to overcome this problem by recasting the decoding problem in a multi-task learning (MTL) framework. In MTL, a single PR model is used to learn different but related “tasks” simultaneously. The primary advantage of MTL is that it makes more efficient use of the data available and leads to more accurate models by making use of the relationships between tasks. In this work, we construct MTL models where each subject is modelled by a separate task. We use a flexible covariance structure to model the relationships between tasks and induce coupling between them using Gaussian process priors. We present an MTL method for classification problems and demonstrate a novel mapping method suitable for PR models. We apply these MTL approaches to classifying many different contrasts in a publicly available fMRI dataset and show that the proposed MTL methods produce higher decoding accuracy and more consistent discriminative activity patterns than currently used techniques. Our results demonstrate that MTL provides a promising method for multi-subject decoding studies by focusing on the commonalities between a group of subjects rather than the idiosyncratic properties of different subjects.

Highlights

Pattern recognition (PR) methods are becoming increasingly important tools for neuroimaging data analysis and are complementary to more conventional mass-univariate analysis methods based on the general linear model (GLM; Friston et al (1995))
We demonstrated a translational application of multi-task learning (MTL) for fMRI data analysis
We evaluated several different MTL approaches based on Gaussian process (GP) which enabled the automatic estimation of the covariance structure between fMRI data from different subjects

Summary

Introduction

Pattern recognition (PR) methods are becoming increasingly important tools for neuroimaging data analysis and are complementary to more conventional mass-univariate analysis methods based on the general linear model (GLM; Friston et al (1995)). Mass-univariate methods, or encoding models (Naselaris et al, 2011), are well suited to mapping focal, group level associations between experimental variables and brain structure or function. PR methods, or decoding models, aim to make predictions based on the spatial or spatiotemporal pattern within the data. PR methods have been useful for making predictions at the single subject level in clinical research studies (Orru et al, 2012) and for detecting neural activity patterns characteristic of instantaneous cognitive states (Norman et al, 2006). Neuroimaging data are well-known to be characterised by substantial inter-subject variability, due to a range of factors including residual registration error, variations in inter-subject functional anatomy (Frost and Goebel, 2012; Morosan et al, 2001) and individual variations in the haemodynamic response (Aguirre et al, 1998; Handwerker et al, 2004). In a mass-univariate context, this variability has been historically

Objectives

Methods

Results

Conclusion