Brain structure carries clues about who prefers instant gratification, study suggests

A brain imaging study of 146 young adults suggests that machine learning models can partly predict how strongly a person favors immediate rewards over larger future ones, using only the volumes of specific brain regions. Regions involved in reward processing and long-term planning featured prominently in the predictions. The findings, published in JAMA Psychiatry, offer early proof-of-concept evidence that brain structure carries information about impulsive decision-making.

In everyday life, people regularly weigh smaller, immediate rewards against larger benefits that require waiting, such as spending money right away versus saving for the future. Psychologists refer to the tendency to reduce the subjective value of a reward the longer one has to wait for it as delay discounting. A person who prefers $20 today over $50 in a month exhibits steep delay discounting, which serves as a behavioral measure of impulsivity and short-sighted decision-making.

Patterns of delay discounting vary widely from person to person and appear at extreme ends of the spectrum across several mental health conditions. Steep discounting is characteristic of disorders involving impulse-control challenges, including substance use disorders, attention-deficit/hyperactivity disorder (ADHD), bulimia nervosa, and borderline personality disorder. By contrast, unusually low discounting, which indicates a strong preference for delayed rewards over immediate ones, is observed in people with anorexia nervosa and in older adults with higher suicidal tendencies.

Because this decision-making trait cuts across different diagnoses, it aligns with the Research Domain Criteria (RDoC) framework, a psychiatric research model that examines mental health along continuous dimensions of behavior and biology rather than strict diagnostic categories. Research suggests that discounting can be shifted in experiments, which makes it a potential target for treatment. Discounting also has practical stakes: a neuroimaging study of adult smokers found that those with steeper discounting were more likely to relapse after smoking-cessation treatment, and showed differences in brain activity in regions tied to decision-making and habit formation.

Most brain imaging research on impulsive choice has focused on functional activity or communication between brain networks during decision-making tasks. Along these lines, a brain imaging study of adolescents found that stronger moment-to-moment coupling between the amygdala, an emotional hub, and the brain’s cognitive control network predicted a greater preference for immediate rewards, but only among 14- to 17-year-olds making choices about high-value rewards. However, it remains unknown whether the brain’s physical structure, which changes far more slowly than its activity, can reliably indicate a person’s typical level of delay discounting.

Free daily newsletter

Physical brain tissue is broadly divided into two main components that make up the brain’s macrostructural, or large-scale, anatomy: gray matter and white matter. Gray matter consists primarily of neuronal cell bodies located in the brain’s folded outer layer and deep internal structures that process information, whereas white matter consists of bundles of insulated nerve fibers beneath the surface that connect different gray matter areas to one another.

To test whether machine learning could detect signals of individual delay discounting from the volume of these two tissue types, co-first authors Bryan Youngwoo Yoon of Columbia University Irving Medical Center and Wonjin Kim of Yonsei University conducted an initial proof-of-concept study alongside senior authors Joon Yul Choi of Yonsei University and Wi Hoon Jung, an associate professor of psychology at Gachon University in Seongnam, South Korea.

“Despite growing interest in the neural mechanisms underlying impulsive decision-making, we still know relatively little about how individual differences in delay discounting are reflected in the brain’s macrostructural anatomy,” Jung told PsyPost. “First, although previous research has highlighted the involvement of distributed brain networks in decision-making, the specific contributions of regional gray matter and white matter volumes remain insufficiently understood. Second, we wanted to determine whether machine learning could identify meaningful structural patterns from conventional structural MRI data and use them to predict individual differences in delay discounting. Our goal was to explore whether brain structure could provide useful information about how people balance immediate rewards against future benefits.”

The researchers analyzed behavioral and neuroimaging data collected between July 2020 and July 2023 from 146 college-aged participants, who had an average age of about 22.6 years and included 75 females and 71 males. Each participant completed a behavioral delay discounting assessment and underwent a T1-weighted magnetic resonance imaging (MRI) scan, a standard structural scan that uses magnetic fields and radio waves to generate detailed three-dimensional images of brain anatomy.

Using a standardized anatomical atlas, the team extracted the volumes of gray matter and white matter across predefined brain subdivisions, known as regions of interest. To control for natural differences in overall head size across participants, every regional volume was normalized to the person’s total intracranial volume, which represents the total space inside the skull cavity.

To train and test their predictive algorithms, the authors generated five independent random splits of the dataset, placing 80% of the participants into a training set and holding out the remaining 20% as a testing set. Within each 80% training set, the team used generalized linear models, a statistical method for estimating relationships between variables, to identify brain regions whose volumes were statistically associated with delay discounting after adjusting for age and sex assigned at birth. The researchers then ran an exhaustive subset search across those candidate regions, systematically testing every possible combination of brain areas to find the specific subset that best estimated delay discounting.

The optimal combinations of gray matter or white matter regions were fed into a multilayer perceptron model to predict delay discounting scores in the held-out 20% test sets. A multilayer perceptron is a type of artificial neural network used in machine learning that passes input data through multiple interconnected layers of mathematical nodes, allowing it to detect complex patterns across multiple brain regions at once. The researchers built and evaluated separate neural network models for white matter volumes and gray matter volumes.

Across the five testing splits, both the white matter model and the gray matter model estimated individual delay discounting scores with moderate accuracy. The white matter model achieved an average R-squared value (the proportion of variance in an outcome accounted for by a model) of 0.22, with a standard deviation of 0.11, indicating that white matter volumes in the selected regions accounted for an average of 22% of the variation in delay discounting. The gray matter model reached an average R-squared value of 0.19 (with a standard deviation of 0.10), accounting for an average of 19% of the variation in delay discounting.

“The predictive performance was moderate but meaningful for a study attempting to estimate a complex behavioral trait from brain structure alone,” Jung explained. “First, these results indicate that regional brain volumes contain information relevant to individual differences in delay discounting. Second, the models did not explain all of the variability, which is unsurprising given that decision-making is influenced by many interacting factors, including brain structure, functional brain activity, genetics, and environmental experiences. We therefore see these findings as evidence that structural MRI can contribute to understanding and predicting impulsive decision-making, rather than as evidence that brain anatomy alone can accurately explain complex behavior.”

When the researchers examined which specific brain structures drove the predictions, they found that two areas in the right hemisphere were selected in both the white matter model and the gray matter model. One was the right accumbens area, a structure deep inside the brain’s ventral striatum that plays a central role in processing reward, pleasure, and motivation. The other was the right frontal pole, situated at the frontmost tip of the frontal lobe, which supports future planning, weighing long-term consequences, and goal-directed decision-making.

The white matter model also drew on four other regions, mostly involved in sensory and visual processing, while the gray matter model added seven more, including frontal areas tied to self-control and parts of the visual cortex. The paper does not report whether larger or smaller volumes in these areas went with steeper discounting.

The authors point out that the most informative structural features consistently mapped onto the frontolimbic circuit, a brain network that connects frontal planning hubs with deep reward centers, while also implicating sensory and temporal areas that are less commonly highlighted in impulse-control research. Because the white matter model outperformed the gray matter model, the researchers suggest that the macroscopic volume of white matter pathways may be just as important as the volume of gray matter processing hubs in regulating impulsive choices. In the authors’ reading, the accumbens area and frontal pole could represent structural targets for psychiatric conditions characterized by impaired impulse control.

“The main message is that differences in how people weigh immediate rewards against future benefits may be reflected, at least in part, in the structure of their brains,” Jung said. “These findings suggest that individual differences in impulsive decision-making may be associated with structural characteristics of brain regions involved in balancing immediate rewards with longer-term goals.”

Jung noted that one unexpected aspect of the results was that the white matter model outperformed the gray matter model. “Much of the neuroimaging literature on impulsivity has focused on gray matter regions involved in reward processing and cognitive control,” he told PsyPost. “Our results suggest that white matter volume may also provide valuable information about individual differences in delay discounting.”

“This was an interesting finding because it highlights the potential importance of considering both tissue types when investigating the structural basis of decision-making,” he added. “At the same time, because our white matter measures reflected regional tissue volumes rather than direct measures of fiber integrity, further research will be needed to clarify the underlying mechanisms.”

There are a few things to keep in mind when interpreting these findings. The authors framed their work as an initial proof-of-concept study, and the analysis relied on a modest sample of 146 college-aged young adults scanned using a single MRI scanner. Although the machine learning models were evaluated on held-out portions of the dataset, they have not yet been tested on an independent external sample or across diverse age groups and clinical populations.

In addition, the models accounted for an average of 19% to 22% of the variation in delay discounting, which the authors describe as in line with the medium effect sizes typical of brain-based predictions of behavior. That still leaves most individual differences unexplained. Each test set included only about 29 people, and accuracy varied considerably from split to split, so the estimates could shift in a larger sample.

The cross-sectional nature of the study also means that brain anatomy and delay discounting were measured at a single point in time. Because of this design, the results cannot establish whether structural volume differences in the accumbens area or frontal pole cause people to favor immediate rewards, or whether long-standing behavioral habits and environmental factors gradually shape brain tissue volume. At the same time, the study relied solely on T1-weighted structural MRI scans, meaning the white matter measurements reflect bulk tissue volume within or next to defined brain regions rather than the microscopic integrity or organization of nerve fiber bundles.

“These measures do not directly assess the microstructural integrity of white matter fibers, as diffusion MRI measures such as fractional anisotropy or tractography-based analyses attempt to do,” Jung pointed out. “Future studies combining these approaches may provide a more comprehensive understanding of the structural basis of delay discounting.” In neuroimaging, diffusion MRI methods such as fractional anisotropy and tractography track the movement of water molecules along nerve fibers to map their microscopic alignment and wiring pathways.

“Our findings should not be interpreted as suggesting that brain scans can determine how an individual will behave,” Jung added. “The models identify statistical patterns associated with delay discounting, not fixed biological traits that dictate behavior. Delay discounting can vary with circumstances and may change through learning or intervention. Our findings therefore point to potential neural correlates of individual differences, rather than a deterministic explanation of impulsivity.”

To expand on this work, the researchers suggest validating their models in larger, external cohorts across multiple research sites and integrating different types of brain imaging data. Future studies could combine structural and diffusion scans with resting-state functional connectivity, which measures how brain regions coordinate their activity while a person is at rest. Combining these approaches could help scientists determine whether structural brain markers of delay discounting can aid in early diagnosis or inform personalized interventions for psychiatric disorders.

“Combining structural T1-weighted MRI with diffusion MRI measures and resting-state functional connectivity could provide a more comprehensive picture of how brain anatomy and connectivity relate to delay discounting,” Jung said. “Such an approach may also help explain why white matter volumes showed stronger predictive performance than gray matter volumes in the present study.”

The team also hopes to test these neuroimaging models in clinical groups. By uniting computational brain imaging with behavioral economics, a field that studies how psychological factors shape financial and personal choices, researchers aim to better understand impulse-control challenges across different diagnoses.

“We are interested in extending this work to clinical populations, including individuals with substance use disorders, eating disorders, and ADHD,” he said. “By examining whether the identified structural patterns are also relevant in these groups, we hope to determine whether neuroimaging-based measures of delay discounting could eventually contribute to identifying behavioral risk profiles or monitoring responses to targeted interventions. However, their clinical usefulness would need to be established through careful external validation and prospective studies.”

“One aspect of this study that we find particularly valuable is the opportunity to bring together behavioral economics and computational neuroimaging to investigate a clinically relevant dimension of behavior,” Jung concluded. “Rather than treating impulsivity solely as a symptom associated with a particular diagnosis, we approached delay discounting as a continuous behavioral characteristic within the RDoC framework.”

“We believe this dimensional perspective may help researchers better understand why people differ in their decision-making, both within and across psychiatric diagnoses. Although our findings represent an early step rather than an immediate clinical application, they demonstrate the potential of combining theoretically grounded behavioral measures with machine learning to investigate the relationship between brain structure and complex human behavior.”

The study, “Structural Brain Volumes and Delay Discounting,” was authored by Bryan Youngwoo Yoon, Wonjin Kim, Dohyeon Kim, Joon Yul Choi, and Wi Hoon Jung.