A decade of decoding reward-related fMRI signals and where we go from here
Introduction
Approaching potential rewards in the environment while avoiding danger is critical for survival. To enable such adaptive behaviors, nervous systems of humans and other animals must process and represent reward-related information about future and experienced outcomes. In turn, disruptions in neural reward processing may result in maladaptive behaviors such as those observed in patients with neurological and psychiatric disorders.
Neuroscience has a rich history of studying reward processing in animals using lesions (Everitt and Stacey, 1987, Gaffan and Murray, 1990, Holland and Gallagher, 1993, Schoenbaum et al., 2003), electrophysiology (Romo and Schultz, 1990, Schoenbaum et al., 1998, Tremblay and Schultz, 1999) and, more recently, optogenetics (Ferenczi et al., 2016, Gremel and Costa, 2013, Steinberg et al., 2013). In contrast, the study of human reward processing has traditionally been limited to evidence from patients with lesions in circumscribed brain areas (Bechara et al., 1994, Tsuchida et al., 2010). This drastically changed with the invention of functional magnetic resonance imaging (fMRI), which has opened a window into the neural mechanisms of reward in humans (Breiter et al., 2001, Knutson et al., 2001, O’Doherty et al., 2001). Although univariate fMRI studies have taught us a great deal about the brain regions involved in reward processing (Bartra et al., 2013, Clithero and Rangel, 2013), they are restricted to brain areas in which neuronal activity is relatively uniformly associated with reward across voxels and individuals. The advent of multivariate approaches to fMRI data analysis (Haxby et al., 2001, Haynes and Rees, 2005, Kamitani and Tong, 2005, Kriegeskorte et al., 2006) mitigated these limitations and enabled us to study brain regions in which neurons represent reward-related information heterogeneously across space and individuals.
Over the past decade, fMRI studies using MVPA techniques have decoded a wide array of reward-related signals from the human brain. Individual studies have examined feedback-related signals predicting value-based choices (Hampton and O’Doherty, 2007), and how reinforcement signals from the outcome of simple games are represented (Vickery et al., 2011). Other studies have focused on decoding the subjective value of offers (Krajbich et al., 2009, Wang et al., 2014), the value of predicted outcomes (Kahnt et al., 2010, Kahnt et al., 2011b), consumer choices made inside the scanner (Grosenick et al., 2008), and differences between types of valuation and reward (Clithero et al., 2009, Clithero et al., 2011). Moreover, several studies used brain activity acquired during passive viewing (Levy et al., 2011), or even outside the focus of attention (Tusche et al., 2010, Tusche et al., 2013), to predict preference-based choices made outside the scanner (Smith et al., 2014). For instance, Tusche et al. (2010) presented a series of images of automobiles in the background while subjects were engaged in a demanding visual fixation task. The acquired fMRI data were then used to predict hypothetical consumer choices made after scanning. Similarly, Levy et al. (2011) showed that preference-related fMRI signals during passive viewing of images of consumer products can be used to predict actual choices at a later time point. Beyond demonstrating the technical feasibility of decoding real-world decisions based on fMRI signals, these studies have illustrated potential applications of reward-based decoding approaches to infer choice and valuation when open behavior is unavailable (Grosenick et al., 2008, Smith et al., 2014).
In general, MVPA studies in the reward domain have confirmed previous results from univariate fMRI experiments and successfully decoded reward-related information in subcortical areas such as the striatum, and in areas of the prefrontal and parietal cortex. Section 2 of this article discusses common models and assumptions that relate MVPA findings to the reward-related firing of single neurons. (Note that no general overview on technical aspects of fMRI decoding methods and multivariate classifiers will be provided, and the interested reader is referred to comprehensive review articles on this topic (Haynes, 2015, Misaki et al., 2010, Mourao-Miranda et al., 2005).) Beyond mapping reward-related brain regions, many MVPA studies have addressed additional questions regarding neural mechanisms of reward that would have been difficult to address with conventional imaging approaches. A selection of such studies is the focus of section 3 wherein key points arising from these experiments are summarized. Section 4 discusses how representational imaging methods are being used more recently to address questions related to reward and goal-directed behavior that go beyond the encoding of value.
Section snippets
Reward coding in single neurons
How do neuronal populations encode information about reward? Single- and multi-unit recordings in animals indicate that depending on the brain region, the relationship between reward parameters and neuronal firing rates is either simple and homogeneous, or complex and heterogeneous. For instance, dopaminergic neurons in the substantia nigra and ventral tegmental area display a relatively homogeneous coding scheme, such that firing rates increase with the value of unpredicted rewards and
What have we learned from decoding reward signals?
In general, reward studies that use MVPA approaches have largely confirmed previous results from univariate fMRI studies. The following section focuses on a selection of studies that have used MVPA methods to answer questions related to reward processing that go beyond what is known from, and what can be typically achieved by, traditional univariate analysis methods.
Current topics in decoding reward
This section draws from studies that highlight the potential of MVPA methods to ask more detailed questions about the nature of reward representations. Specifically, it reviews studies that use MVPA to demonstrate how reward signals can take the form of a common currency for value, and how these general value signals contrast with highly specific reward representations which simultaneously encode multiple features of expected outcomes that are not necessarily related to value.
Discussion and conclusions
Multivariate decoding techniques have offered new and exciting ways to analyze fMRI data, and they have substantially extended the scope of questions that can be addressed. The study of human reward learning and decision-making has benefitted from this advance. The main takeaway from reward studies using MVPA methods is that in many regions of the brain, encoding of reward information is not limited to value. Specifically, reward predictive signals, especially in prefrontal cortex, incorporate
Acknowledgments
The author thanks Drs. J.D. Howard, L.P. Qu, and P.N. Tobler for insightful comments and suggestions. The author is supported by grants from the National Institute on Deafness and Other Communication Disorders (NIDCD) and the National Institute on Drug Abuse (NIDA), National Institutes of Health, USA.
References (129)
- et al.
The valuation system: a coordinate-based meta-analysis of BOLD fMRI experiments examining neural correlates of subjective value
Neuroimage
(2013) - et al.
Insensitivity to future consequences following damage to human prefrontal cortex
Cognition
(1994) - et al.
Two anatomically and computationally distinct learning signals predict changes to stimulus-outcome associations in hippocampus
Neuron
(2016) - et al.
Functional imaging of neural responses to expectancy and experience of monetary gains and losses
Neuron
(2001) - et al.
Dopamine in motivational control: rewarding, aversive, and alerting
Neuron
(2010) - et al.
Encoding the identity and location of objects in human LOC
Neuroimage
(2011) - et al.
Local pattern classification differentiates processes of economic valuation
Neuroimage
(2009) - et al.
Within- and cross-participant classifiers reveal different neural coding of information
Neuroimage
(2011) Is cortical vasculature functionally organized?
Neuroimage
(2010)- et al.
Multi-scale classification of disease using structural MRI and wavelet transform
Neuroimage
(2012)
A primer on pattern-based approaches to fMRI: principles, pitfalls, and perspectives
Neuron
Reading hidden intentions in the human brain
Curr. Biol.
Perceptual learning and decision-making in human medial frontal cortex
Neuron
Decoding different roles for vmPFC and dlPFC in multi-attribute decision making
Neuroimage
Social information signaling by neurons in primate striatum
Curr. Biol.
A region of mesial prefrontal cortex tracks monetarily rewarding outcomes: characterization with rapid event-related fMRI
Neuroimage
How does an fMRI voxel sample the neuronal activity pattern: compact-kernel or complex spatiotemporal filter?
Neuroimage
Dynamic interaction between reinforcement learning and attention in multidimensional environments
Neuron
The root of all value: a neural common currency for choice
Curr. Opin. Neurobiol.
Comparison of multivariate classifiers and response normalizations for pattern-information fMRI
Neuroimage
Neural economics and the biological substrates of valuation
Neuron
Classifying brain states and determining the discriminating activation patterns: support Vector Machine on functional MRI data
Neuroimage
Encoding and decoding in fMRI
Neuroimage
Bayesian reconstruction of natural images from human brain activity
Neuron
Multivariate representation of food preferences in the human brain
Brain Cogn.
Expectation modulates neural representations of valence throughout the human brain
Neuroimage
fMRI orientation decoding in V1 does not require global maps or globally coherent orientation stimuli
Front. Psychol.
Repetition suppression: a means to index neural representations using BOLD?
Philos. Trans. R. Soc. Lond. B Biol. Sci.
The role of the orbitofrontal cortex in the pursuit of happiness and more specific rewards
Nature
A distinct role of the temporal-parietal junction in predicting socially guided decisions
Science
Autocorrelation structure at rest predicts value correlates of single neurons during reward-guided choice
eLife
A probability distribution over latent causes, in the orbitofrontal cortex
J. Neurosci.
A sensitive and specific neural signature for picture-induced negative affect
PLoS Biol.
Evidence for a common representation of decision values for dissimilar goods in human ventromedial prefrontal cortex
J. Neurosci.
Population coding of affect across stimuli, modalities and individuals
Nat. Neurosci.
Resolving human object recognition in space and time
Nat. Neurosci.
Informatic parcellation of the network involved in the computation of subjective value
Soc. Cogn. Affect. Neurosci.
Measuring neural representations with fMRI: practices and pitfalls
Ann. N. Y. Acad. Sci.
The neuro-computational architecture of value-based selection in the human brain
Cereb. Cortex
Single-unit recordings in the macaque face patch system reveal limitations of fMRI MVPA
J. Neurosci.
Studies of instrumental behavior with sexual reinforcement in male rats (Rattus norvegicus): ii. Effects of preoptic area lesions, castration, and testosterone
J. Comp. Psychol.
Prefrontal cortical regulation of brainwide circuit dynamics and reward-related behavior
Science
Action-specific value signals in reward-related regions of the human brain
J. Neurosci.
Coarse-scale biases for spirals and orientation in human visual cortex
J. Neurosci.
Amygdalar interaction with the mediodorsal nucleus of the thalamus and the ventromedial prefrontal cortex in stimulus-reward associative learning in the monkey
J. Neurosci.
The Human Connectome project’s neuroimaging approach
Nat. Neurosci.
Human orbitofrontal cortex mediates extinction learning while accessing conditioned representations of value
Nat. Neurosci.
Encoding predictive reward value in human amygdala and orbitofrontal cortex
Science
Orbitofrontal and striatal circuits dynamically encode the shift between goal-directed and habitual actions
Nat. Commun.
Interpretable classifiers for FMRI improve prediction of purchases
IEEE Trans. Neural Syst. Rehabil. Eng.
Cited by (45)
Spatiotemporal Precision of Neuroimaging in Psychiatry
2023, Biological PsychiatryCitation Excerpt :Neuroimaging can complement such computational models of decision making in psychopathology by measuring a reward prediction error signal (i.e., the difference between the reward that was received and the reward that was expected), a key computational component in reinforcement learning and active inference models (101). Reward prediction error signals localize to specific neurochemical circuitry (e.g., dopaminergic pathways) and are observable in both MEG/EEG (102,103) and fMRI (104). Reward prediction error signals, detected with fMRI, accurately predict response to CBT in depression, where an increased responsivity of the amygdala and striatum to unexpected rewards has been interpreted as indicating susceptibility to subsequent belief updating during cognitive restructuring during CBT (105).
More complex than you might think: Neural representations of food reward value in obesity
2022, AppetiteCitation Excerpt :Another relevant factor is that most previous studies used only mass-univariate analyses for their fMRI-data, which only inform on where in the brain the level of brain activity is relatively high. However, empirical studies have shown that the level of brain activity in the mesocorticolimbic system is actually unlikely to reflect the reward value of the presented visual food stimuli (Chikazoe et al., 2014a; Janssen et al., 2019; Kahnt, 2018; Salamone & Correa, 2012; Suzuki et al., 2017). Brain activity in this system, instead, may reflect salience.
Testing the distributed representation hypothesis in object recognition in two open datasets
2022, Neuroscience Letters