Author name cluster

Tor Wager

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

7 papers

1 author row

YNIMG Journal 2023 Journal Article

A machine learning based approach towards high-dimensional mediation analysis

Tanmay Nath
Brian Caffo
Tor Wager
Martin A. Lindquist

Mediation analysis is used to investigate the role of intermediate variables (mediators) that lie in the path between an exposure and an outcome variable. While significant research has focused on developing methods for assessing the influence of mediators on the exposure-outcome relationship, current approaches do not easily extend to settings where the mediator is high-dimensional. These situations are becoming increasingly common with the rapid increase of new applications measuring massive numbers of variables, including brain imaging, genomics, and metabolomics. In this work, we introduce a novel machine learning based method for identifying high dimensional mediators. The proposed algorithm iterates between using a machine learning model to map the high-dimensional mediators onto a lower-dimensional space, and using the predicted values as input in a standard three-variable mediation model. Hence, the machine learning model is trained to maximize the likelihood of the mediation model. Importantly, the proposed algorithm is agnostic to the machine learning model that is used, providing significant flexibility in the types of situations where it can be used. We illustrate the proposed methodology using data from two functional Magnetic Resonance Imaging (fMRI) studies. First, using data from a task-based fMRI study of thermal pain, we combine the proposed algorithm with a deep learning model to detect distributed, network-level brain patterns mediating the relationship between stimulus intensity (temperature) and reported pain at the single trial level. Second, using resting-state fMRI data from the Human Connectome Project, we combine the proposed algorithm with a connectome-based predictive modeling approach to determine brain functional connectivity measures that mediate the relationship between fluid intelligence and working memory accuracy. In both cases, our multivariate mediation model links exposure variables (thermal pain or fluid intelligence), high dimensional brain measures (single-trial brain activation maps or resting-state brain connectivity) and behavioral outcomes (pain report or working memory accuracy) into a single unified model. Using the proposed approach, we are able to identify brain-based measures that simultaneously encode the exposure variable and correlate with the behavioral outcome.

Details DOI

RLDM Conference 2015 Conference Abstract

Self-reinforcing expectancy effects on pain: Behavioral and brain mechanisms

Marieke Jepma
Tor Wager

Cues associated with pain or pain relief through classical conditioning can profoundly mod- ify responses to subsequent noxious events. Unlike conditioned fear, conditioned pain modulation can be resistant to extinction or even grow over time in the absence of reinforcement. One explanation for such ‘self-reinforcing’ effects is that prior beliefs bias learning, by enhancing learning from expectancy- confirming relative to disconfirming events. If so, there is the potential for positive feedback loops between expectations and experiences that create ‘self-fulfilling prophecies. ’ In two experiments (N=26 and 34), we examined the behavioral and brain mechanisms underlying interactions between pain expectations and pain experiences. Participants first completed a conditioning phase, in which cues were repeatedly paired with either low or high heat levels. In a subsequent ‘extinction’ phase, all cues were followed by iden- tical noxious heat, and we measured trial-to-trial dynamics in expectations, pain experience, autonomic responses, and (in Experiment 2) fMRI activity. Subjective, autonomic and neural responses to painful events were stronger following high- than low-pain cues, and these effects were mediated by self-reported pain expectancies. These effects did not extinguish over time. Analyses of learning dynamics revealed that participants updated pain expectancies more following outcomes that confirmed expectations than those that disconfirmed them, indicating a confirmation bias that maintained the cues’ effects on pain in the absence of reinforcement. Individual differences in the strength of this confirmation bias correlated with anterior cingulate cortex activation to expectancy-confirming vs. -disconfirming outcomes, suggesting a key role for the cingulate in the regulation of learning rate as a function of prior beliefs. These results can help explain why beliefs in many domains can have persistent effects even in the absence of confirming evidence.

PDF Details

RLDM Conference 2013 Conference Abstract

Dissociations in reward network activation during informative and affective feedback

Jenna Reinen
Catherine Insel
Tor Wager
Daphna Shohamy

Converging evidence across species and methods have indicated that a specific network of brain regions supports incremental, trial-by-trial learning from feedback. Functional imaging (fMRI) has provided evidence that in humans, a learning signal known as prediction error (PE), is represented in these regions including the striatum, orbitofrontal cortex (OFC), and cingulate. More recent data has also shown PE in the medial temporal lobe (MTL), which has implications for value-based decision making across mem- ory systems. Importantly, these signals are thought to support the ability of an organism to update value associated with a cue in a dynamic environment, and to drive motivated behavior and decision making. However, the ability to perform optimally entails both experiencing an intact, appropriate hedonic response to a reward in order to assess value, as well as being able to associate this value-based information with the stimulus. To date, most studies collapse information about a reward and reward-related affective expe- rience together, which does not allow one to identify the separate neural contributions of these regions to these qualitatively different types of feedback. To address this, we tested 25 subjects on a two-stage, proba- bilistic feedback-based learning task while undergoing fMRI. Participants made choices during two phases of non-intermixed motivational conditions (gain, loss). On each trial, subjects chose between two shapes and received stochastically-delivered informative feedback (correct or incorrect) followed by an uncertain amount of hedonic feedback (monetary gain or loss). Results indicated that there was a dissociation in sever- al brain structures, including the OFC and MTL, related to PE in different motivational contexts, and when receiving different types of feedback. These findings suggest that certain structures in the reward network are sensitive to motivational context and affective experience during feedback.

PDF Details

RLDM Conference 2013 Conference Abstract

Dynamic representations of pain anticipation

Marieke Jepma
Matt Jones
Tal Yarkoni
Tor Wager

Understanding the neural computations underlying how expectancies are developed has received surprisingly little attention, despite their central role in theories of learning, behavior, and value. Preliminary evidence in the context of pain suggests that brain regions involved in anticipating pain (e. g. , ACC, Insula, SII) overlap with those involved in its experience. However, it remains an open question if these regions are actually representing the magnitude of the predicted pain and if they dynamically adapt to changes in the perception of anticipated pain. To address this question we developed a novel method of estimating model parameters by treating brain regions as learners and directly fitting reinforcement learning models to fMRI data. Twenty-eight participants learned associations between visual cues and different levels of heat pain applied to their left forearm. Each cue was associated with a different distribution of noxious heat intensity. We represented the value of pain as a linear increase that begins ramping up at the onset of the cue and stops at pain onset with the magnitude of the pain value function determined using a simple Rescorla-Wagner pre- diction error algorithm. We fit the model directly to the brain data and used a model comparison procedure to identify voxels that fit the data significantly better than a competing null model that did not learn. The results revealed that a distributed set of regions including somatosensory cortex, SII, bilateral amygdala, and the dorsal ACC are involved in anticipating the magnitude of future pain with greater activation in these re- gions being associated with larger predictions of pain magnitude. Interestingly, the vmPFC which has been widely implicated in processing reward value decreases as greater pain is anticipated. These results suggest that pain anticipation involves integrating information from both sensory and affective systems.

PDF Details

RLDM Conference 2013 Conference Abstract

Mind matters: Placebo enhances reward learning in Parkinson’s disease

Liane Schmidt
Erin Braun
Tor Wager
Daphna Shohamy

Parkinson’s disease (PD) is characterized by a loss of midbrain dopamine neurons that play a central role in reward learning. Dopaminergic drugs can restore dopamine and improve reward-learning deficits in PD patients. Interestingly, a combination of the expectation of relief associated with taking a drug and conditioned drug responses have been shown to trigger endogenous dopamine release in the brain. The functional significance of this placebo effect is, however, unclear. Here we addressed this question by using functional magnetic resonance imaging (fMRI) to measure the effects of placebo on brain activation in PD patients while they performed an instrumental learning task. To disentangle psychological and pharmacolog- ical effects of the drug, patients were scanned under three conditions: no treatment (off drug), placebo, and levodopa (on drug). Compared to no treatment, placebo and levodopa both enhanced learning from reward. FMRI revealed that this finding was related to enhanced value representation in the ventromedian prefrontal cortex at the time of choice under placebo, as well as when patients were on levodopa, compared with off drug. These findings suggest that the psychological effects of a drug may be, in some cases, as powerful in improving reward learning as the pharmacological effects of a dopamine precursor, and are consistent with findings that placebo can lead to enhanced dopaminergic activity.

PDF Details

YNIMG Journal 2005 Journal Article

Valid conjunction inference with the minimum statistic

Thomas Nichols
Matthew Brett
Jesper Andersson
Tor Wager
Jean-Baptiste Poline

In logic a conjunction is defined as an AND between truth statements. In neuroimaging, investigators may look for brain areas activated by task A AND by task B, or a conjunction of tasks (Price, C. J. , Friston, K. J. , 1997. Cognitive conjunction: a new approach to brain activation experiments. NeuroImage 5, 261–270). Friston et al. (Friston, K. , Holmes, A. , Price, C. , Büchel, C. , Worsley, K. , 1999. Multisubject fMRI studies and conjunction analyses. NeuroImage 10, 85–396) introduced a minimum statistic test for conjunction. We refer to this method as the minimum statistic compared to the global null (MS/GN). The MS/GN is implemented in SPM2 and SPM99 software, and has been widely used as a test of conjunction. However, we assert that it does not have the correct null hypothesis for a test of logical AND, and further, this has led to confusion in the neuroimaging community. In this paper, we define a conjunction and explain the problem with the MS/GN test as a conjunction method. We present a survey of recent practice in neuroimaging which reveals that the MS/GN test is very often misinterpreted as evidence of a logical AND. We show that a correct test for a logical AND requires that all the comparisons in the conjunction are individually significant. This result holds even if the comparisons are not independent. We suggest that the revised test proposed here is the appropriate means for conjunction inference in neuroimaging.

Details DOI

YNIMG Journal 2002 Journal Article

Functional Neuroanatomy of Emotion: A Meta-Analysis of Emotion Activation Studies in PET and fMRI

K.Luan Phan
Tor Wager
Stephan F. Taylor
Israel Liberzon

Neuroimagingstudies with positron emission tomography (PET) and functional magnetic resonance imaging (fMRI) have begun to describe the functional neuroanatomy of emotion. Taken separately, specific studies vary in task dimensions and in type(s) of emotion studied and are limited by statistical power and sensitivity. By examining findings across studies, we sought to determine if common or segregated patterns of activations exist across various emotional tasks. We reviewed 55 PET and fMRI activation studies (yielding 761 individual peaks) which investigated emotion in healthy subjects. Peak activation coordinates were transformed into a standard space and plotted onto canonical 3-D brain renderings. We divided the brain into 20 nonoverlapping regions, and characterized each region by its responsiveness across individual emotions (positive, negative, happiness, fear, anger, sadness, disgust), to different induction methods (visual, auditory, recall/imagery), and in emotional tasks with and without cognitive demand. Our review yielded the following summary observations: (1) The medial prefrontal cortex had a general role in emotional processing; (2) fear specifically engaged the amygdala; (3) sadness was associated with activity in the subcallosal cingulate; (4) emotional induction by visual stimuli activated the occipital cortex and the amygdala; (5) induction by emotional recall/imagery recruited the anterior cingulate and insula; (6) emotional tasks with cognitive demand also involved the anterior cingulate and insula. This review provides a critical comparison of findings across individual studies and suggests that separate brain regions are involved in different aspects of emotion.

Details DOI