Impulsivity and higher preference for sooner over later rewards (i.e., delay discounting) are transdiagnostic markers of many psychiatric and neurodegenerative disorders. Yet, their neurobiological basis is still debated. Here, we aimed at 1) identifying a structural MRI signature of delay discounting in healthy adults, and 2) validating it in patients with behavioral variant frontotemporal dementia (bvFTD)—a neurodegenerative disease characterized by high impulsivity. We used a machine-learning algorithm to predict individual differences in delay discounting rates based on whole-brain grey matter density maps in healthy male adults (Study 1, N = 117). This resulted in a cross-validated prediction-outcome correlation of r = 0.35 (p = 0.0028). We tested the validity of this brain signature in an independent sample of 166 healthy adults (Study 2) and its clinical relevance in 24 bvFTD patients and 18 matched controls (Study 3). In Study 2, responses of the brain signature did not correlate significantly with discounting rates, but in both Studies 1 and 2, they correlated with psychometric measures of trait urgency—a measure of impulsivity. In Study 3, brain-based predictions correlated with discounting rates, separated bvFTD patients from controls with 81% accuracy, and were associated with the severity of disinhibition among patients. Our results suggest a new structural brain pattern—the Structural Impulsivity Signature (SIS)—which predicts individual differences in impulsivity from whole-brain structure, albeit with small-to-moderate effect sizes. It provides a new brain target that can be tested in future studies to assess its diagnostic value in bvFTD and other neurodegenerative and psychiatric conditions characterized by high impulsivity.
Article
A structural MRI marker predicts individual differences in impulsivity and classifies patients with behavioral-variant frontotemporal dementia from matched controls
https://doi.org/10.21203/rs.3.rs-4794608/v1
This work is licensed under a CC BY 4.0 License
Version 1
posted
You are reading this latest preprint version
brain signature
machine-learning
dementia
decision-making
delay discounting
intertemporal choice
prediction
Impulsivity is the tendency to act in a rush and to seek immediate rewards without consideration of potentially negative long-term consequences 1. Trait impulsivity varies substantially within the general population, with high impulsivity being a hallmark of many psychiatric and neurological conditions 2. Despite the many negative consequences of high impulsivity for health and life in general 3,4, its neurobiological correlates are still unclear, and it is unknown whether individual differences in impulsivity can be reliably predicted based on structural brain features 5–7. Neurobiological measures of impulsivity could help to understand the mechanisms and disentangle the heterogeneity of symptoms related to maladaptive behavior and decision-making. Brain signatures of impulsivity could also constitute new targets for diagnosis and treatment. They might aid in the diagnosis and monitoring of conditions such as behavioral variant frontotemporal dementia (bvFTD)—a neurodegenerative disorder characterized by frontal and temporal brain atrophy, with high impulsivity and inappropriate behaviors as core symptoms 8. In this study, we aimed at developing a structural brain signature of individual differences in impulsivity, and tested whether it could accurately classify patients with bvFTD from matched healthy controls.
The idea that any psychological construct would depend on only one or a few isolated brain regions has been more and more challenged. A new paradigm of “brain signatures” (or “neuromarkers”) promoting a multivariate brain patterns view has therefore emerged, to complement the traditional univariate brain mapping approach examining brain regions independently 9. Brain signatures are predictive models of mental events or of individual variables (such as impulsivity) that take into account distributed information across multiple brain systems 10. Brain signatures using structural data are increasingly used in the field of translational neuroimaging, especially for applications in patients with neurodegenerative conditions 11. One of the greatest advantages of these predictive models which predict behavior from brain features is that they can be tested across studies, labs and populations to challenge their generalizability. We used this brain signature approach to identify a network of spatially distributed structural features associated with impulsivity, as measured by delay discounting. The present study applies the “component process” framework of brain signatures 11. Instead of predicting a given heterogenous condition such as bvFTD, we aimed at identifying a predictive model of a key symptom (i.e., impulsivity), which is a common factor across different diseases. This framework is also suited to the purpose of predicting a specific patient’s clinical profile in a perspective of personalized medicine.
Several arguments support the idea that delay discounting—how much people prefer smaller sooner over larger later rewards—is a reliable measure of stable individual differences in a specific facet of impulsivity (that is the urgency to get short-term rather than long-term reward). Individual differences in delay discounting are relatively stable over time and show significant genetic heritability 12–14. Delay discounting moreover constitutes a potential transdiagnostic marker of conditions with high impulsivity since it has been found to be altered across multiple psychiatric 15 and neurodegenerative conditions 16. Recent studies have therefore started to investigate the neurobiological basis of individual differences in delay discounting 7,14,17–23. However, less is known about how these candidate brain markers of delay discounting are expressed in psychiatric and neurological conditions characterized by increased impulsivity.
Characterized by multiple impulsivity-related symptoms, bvFTD is a good example to demonstrate the clinical potential (in particular for diagnosis) of a structural brain signature of delay discounting. BvFTD is the most common clinical variant of syndromes associated with predominant degeneration of the prefrontal and temporal regions as well as the basal ganglia. It is characterized by significant changes in personality and behavior including disinhibition (socially inappropriate and generally impulsive behaviors), as well as executive function deficits 8. Brain regions known to be related to delay discounting such as the orbitofrontal cortex (OFC), ventromedial prefrontal cortex (vmPFC) and ventral striatum 24–26 are often affected in bvFTD 27,28. Relatedly, most studies found an alteration of delay discounting in bvFTD patients compared to controls 16,29–32.
Here, we first trained and cross-validated a structural MRI-based brain signature in a healthy adult population (Study 1, N = 117) using LASSO-PCR (least absolute shrinkage and selection operator-principal component regression)—an established machine-learning algorithm 33,34 —to predict individual differences in delay discounting rates from subjects’ grey matter maps (N = 117). Brain markers of individual differences need to be tested in different and completely independent samples and studies to establish their robustness and generalizability 10. Thus, in Study 2, we tested the replicability of the brain signature in a second independent sample of healthy adults (N = 166). In Study 3, we tested the validity of the structural brain signature in a clinical population of patients with behavioral variant frontotemporal dementia, who often show high impulsivity and were shown to be steeper discounters (N = 42, including 24 bvFTD patients and 18 matched controls) 35. If a consistent pattern of grey matter density across the brain can reliably predict delay discounting and more generally impulsivity, then the brain-predicted discounting should be higher in bvFTD patients than in controls and should be related to the level of clinically assessed impulsivity in patients. In addition to testing the generalizability of the brain signature developed in Study 1, we analyzed the topographical distribution of the most important structural alterations contributing to differences of brain-predicted delay discounting.
Participants
The research reported here complies with all relevant ethical regulations. The study protocols were approved by the institutional review board of Bonn University’s Medical School (Study 1), by the University of Pennsylvania Institutional Review Board (Study 2), and by the French Ethics Committee “Comité de Protection des Personnes Sud Méditerranée I” (Study 3).
Study 1
In Study 1, participants were recruited in the context of a seven-week dietary intervention study (https://osf.io/rj8sw/?view_only=af9cba7f84064e61b29757f768a8d3bf) at the University of Bonn in Germany. In this study, only male participants were recruited, with the following inclusion criteria: age between 20 and 60 years, right-handedness, non-smoker, no excessive drug or alcohol use in the past year, no psychiatric or neurological disease, body mass index (BMI) between 20 and 34, no other chronic illness or medication, following a typical Western diet without dietary restrictions, and no MRI exclusion criteria (e.g., large tattoos, metal in the body,). For the present purpose, we used only the behavioral and structural MRI data collected during a baseline session before the dietary intervention. N = 117 participants were tested for the baseline session of Study 1. However, four participants were excluded from the present analyses due to being outliers on grey matter density maps (three participants) and due to very incoherent choices at the intertemporal choice task (one participant). Thus, the data of a total of 113 participants was used for the analyses.
Study 2
In Study 2, participants were recruited in the context of a ten-week cognitive training study (registered at clinicaltrials.gov as Clinical trial reg. no. NCT01252966) at the University of Pennsylvania, USA. Individuals between 18 and 35 years of age who reported home computer and internet access were recruited. Exclusion criteria were: an IQ score of < 90 on Shipley Institute of Living Scale, self-reported history of neurological, psychiatric, or addictive disorders (excluding nicotine), positive breath alcohol reading (> 0.01), color blindness, left-handedness, and claustrophobia. Here, we focused on behavioral and structural MRI data collected during the baseline session before the cognitive training. In Study 2, N = 166 participants (mean age = 24.5, 59% male) were included in the baseline session and all were included in our data analyses.
Study 3
For Study 3, participants were recruited in the context of a clinical study at the Paris Brain Institute, France (clinicaltrials.gov: NCT03272230). This study was designed to investigate the behavioral correlates and neural bases of neuropsychiatric symptoms associated with behavioral variant frontotemporal dementia (bvFTD). BvFTD patients were recruited in two tertiary referral centers, at the Pitié-Salpêtrière Hospital and the Lariboisière Fernand-Widal Hospital, in Paris. Patients were diagnosed according to the International Consensus Diagnostic Criteria 8. To be included, bvFTD patients had to present a Mini-Mental State Evaluation (MMSE) score of at least 20. Healthy controls (HC) were recruited by an online announcement. Inclusion criteria included a MMSE score of at least 27 and matching the demographic characteristics of the bvFTD group. In total, 24 bvFTD patients (mean age = 66.6, 66.6% male) and 18 controls matched to patients for age and gender (mean age = 62.6, 44.4% male) were recruited in this clinical study (see Supplementary table 1). Data of all participants were used for our analyses.
Intertemporal choice tasks
Study 1
During the intertemporal choice (ITC) task performed in an MRI scanner, participants in Study 1 were presented with 108 trials offering a choice between a smaller sooner (SS) reward option and a larger later (LL) reward option 14. Participants were informed that one of their choices could be paid out at the end of the experiment, which made their choices non-hypothetical and incentive-compatible. The two options were displayed on the left or right of the screen (position randomized) for 4 seconds. Participants used their left or right index finger to press the response key corresponding to their choice (left index for left option or right index for right option). The option chosen by the participant was then highlighted by a yellow frame which remained on the screen until the end of the 4 second trial. Trials were presented in randomized order (see Koban et al., 2023 for further details on the trial structure).
Study 2
During the ITC performed in an MRI scanner, participants had to make 120 choices between the same smaller immediate reward ($20 today) and a varying larger reward available after a longer delay (e.g., $40 in a month) 36 . Participants were informed that one of their choices could be paid out at the end of the experiment, which made their choices non-hypothetical and incentive-compatible. Each trial started with the presentation of the amount and delay of the larger later option. Once subjects had made their choice, a checkmark on the screen indicated if the larger later option was chosen and a “X” indicated that the immediate option was chosen for 1 s. Subjects had 4 s to make their choice.
Study 3
In Study 3, participants performed two ITC tasks on a computer screen, one using monetary rewards (from 8 to 35 euros) and one using food rewards (from 8 to 35 chocolates) in randomized order 35. In this study, using these two tasks allowed us to test the validity of our brain signature for the prediction of discounting of several types of reward, and thus to investigate generalizability across reward domains. Each of these tasks included 32 choices between SS and LL options. Participants were instructed that one of their 32 choices could be randomly selected and the option that they had chosen would be given to them. Thus, like in Study 1 and 2, participants’ choices were non-hypothetical and incentive-compatible. For each trial, participants could indicate their choice by pressing either a blue key on the keyboard with their right-hand index to select the option on the left or a yellow key with their right-hand middle finger to select the option on the right. Once the choice had been made, a message on the screen indicated which option had been chosen. Trials were presented in randomized order.
Other measures of impulsivity traits and symptoms
Study 1
In Study 1, along with choice data collected from the ITC task, we used self-report data from the Impulsive Behavior Short Scale–8 (I-8), which measures the psychological construct of trait impulsivity according to the Urgency, lack of Premeditation, lack of Perseverance, and Sensation seeking (UPPS) model with four subscales comprising two items each 37. We predicted that the trait of urgency— defined as the tendency to act rashly in an emotional context (e.g., “I sometimes do things to cheer myself up that I later regret”)— would be closest to brain-predicted delay discounting, as both urgency and delay discounting are supposed to measure a tendency to prefer most immediate rewards at the expense of potential long-term gains.
Study 2
In Study 2, we used data from the UPPS-P Impulsive Behavior Scale, which measures trait impulsivity according to the UPPS model with five subscales: positive urgency, negative urgency, lack of premeditation, lack of perseverance, and sensation seeking 38. Paralleling Study 1, we predicted that urgency would be the most closely related to brain-based predictions. We used the average of the subscales of positive urgency (rash actions taken in response to positive emotional states) and negative urgency (rash actions taken in response to negative emotional states) to test this hypothesis.
Study 3
This clinical study did not include a trait measure of impulsivity such as the UPPS scale. However, clinical measures of core symptoms of bvFTD were available, in particular for two symptoms closely related to impulsivity: inhibition deficit and dysexecutive syndrome (i.e., dysfunction in executive functions). In another recent investigation of the same sample, we found that these two bvFTD symptoms are related to higher discounting rates of both money and food 35. We further used the Hayling Sentence Completion Test (HSCT) 39 considered as an objective measure of inhibition deficit, and the Frontal Assessment Battery (FAB) 40 as a measure of executive functions (lower scores indicating worse executive functions). In the HSCT, participants are asked to complete 15 sentences using the appropriate word, as fast as possible (automatic condition, part A), and 15 sentences using a completely unrelated word (inhibition condition, part B). We used the Hayling error score (number of errors in part B) as a measure of the difficulty to inhibit a prepotent response, as in Flanagan et al. 41.
MRI data acquisition and preprocessing
Study 1
Brain imaging data for Study 1 were acquired using a Siemens Trio 3T scanner. Structural images were acquired using a T1 weighted MPRAGE sequence with the following parameters: TR 1660 ms; TE 2.54 ms; FoV 256 mm; 208 slices; slice thickness 0.80 mm; TI 850 ms; flip angle 9°; voxel size 0.8 mm isomorphic; total acquisition time 6:32 min. T1 images were preprocessed for Voxel Based Morphometry (VBM) analyses with SPM 12. We used the SPM module “Segment” to segment and rigidly align T1 images. These images were then used as input into the DARTEL module to create a customized DARTEL template and individual ‘flow fields’ for each subject. DARTEL determines the nonlinear deformations for warping all grey and white matter images so that they match each other. Finally, the SPM module “Normalise to MNI space” generated spatially normalized grey matter images using the deformations estimated in the previous step and images were spatially smoothed with a 6 mm Gaussian FWHM kernel. Among the obtained grey matter images, three outliers (based on Mahalanobis distance of individual grey matter density maps with Bonferroni correction) were detected and excluded from further analyses.
Study 2
Brain imaging data for Study 2 were acquired using a Siemens Trio 3T scanner (with a 32-channel head coil). Structural images were acquired using a T1 weighted MPRAGE sequence with the following parameters: TR 1630 ms; TE 3.11 ms; FOV 192x256; 160 slices; slice thickness 1 mm; TI 1100 ms; flip angle 15°; voxel size 0.9375 × 0.9375 × 1.000 mm; total acquisition time 4:35 min. We used existing data preprocessed by Kable and colleagues 36. T1 images were preprocessed for VBM analyses using the default preprocessing pipeline of the Computational Anatomy Toolbox (CAT12) for SPM12. T1-weighted images underwent denoising filter, were bias corrected, and affine-registered, followed by standard SPM unified tissue segmentation into grey matter, white matter, and cerebral spinal fluid. The grey matter volume images were spatially registered to a common template using Geodesic Shooting, resampled to 1.5 mm3, and spatially smoothed with an 8 mm Gaussian FWHM kernel.
Study 3
Brain imaging data for Study 3 were acquired using a Siemens Prisma whole-body 3T scanner (with a 12-channel head coil). Structural images were acquired using a T1 weighted MPRAGE sequence with the following parameters: TR 2400 ms; TE 2.17 ms; FOV 224 mm; 256 slices; slice thickness 0.70 mm; TI 1000 ms; flip angle 8°; voxel size 0.7 mm isomorphic; total acquisition time 7:38 min. T1 images were preprocessed for Voxel Based Morphometry (VBM) analyses using SPM 12, following the same steps as in Study 1.
Data analyses
The analyses detailed in the following subsections aimed to: (1) develop and validate a structural brain signature predicting delay discounting in a healthy population (Study 1); (2) test the validity of predictions of this structural brain signature as measures of impulsivity in independent studies involving different types of populations, including healthy (Study 2) and clinical samples (Study 3). All analyses were performed using R Studio (1.2.1335) and Matlab (R2017b). The global analytic approach is summarized in Fig. 1A. The specific analyses conducted in each study to check the validity of brain-based predictions are detailed in Fig. 1B.
Computation of discount rates
In all three studies, the individual discounting rate (k) was estimated by fitting logistic regressions to the individual choice data, with the assumption that the subjective value (SV) of the choice options followed hyperbolic discounting, as follows:
where A is the amount of the option, D is the delay until the receipt of the reward (for immediate choice, D = 0), and k is a discounting rate parameter that varies across subjects. Higher values of k indicate greater discounting and thus higher preference for sooner rewards. In Study 1, we used logistic regressions (as described in Wileyto et al., 2004) to estimate the individual parameter k from the participant’s answers in the ICT task at baseline and we used the log(k) values as the parameter to be predicted. Individual k’s were log-transformed in all studies to obtain non-skewed distributions of discounting parameters. In Study 2, we also used the log(k) values at baseline (see 36). In Study 3, we used the log(k) values calculated in bvFTD patients matched with controls for both monetary and food rewards (see 35).
LASSO-PCR, training and cross-validation of the brain pattern predicting log(k) in Study 1
We used a regression-based standard machine learning algorithm, LASSO-PCR (least absolute shrinkage and selection operator-principal component regression) 34, to train a classifier to predict log(k) from the individual whole brain grey matter density (GMD) maps. LASSO-PCR uses principal components analysis (PCA) to reduce the dimensionality of the data and LASSO regression to predict the outcome (log(k)) from the extracted component scores. The components identified by the PCA correspond to groups of brain regions that covary with each other in terms of grey matter density. The LASSO algorithm fits a regularized regression model predicting log(k) from the identified components. This algorithm iteratively shrinks the regression weights towards zero, thus selecting a subset of predictors and reducing the contribution of unstable components. LASSO-PCR is suited to make predictions from thousands of voxels across the whole-brain, in particular because it solves the issue of multicollinearity between voxels and brain regions (see 43,44). Moreover, it is possible to reconstruct voxel weights across the brain (from voxel loadings on PCA components and LASSO regression coefficients of components), yielding predictive brain maps that are easier to interpret than component weights. To assess the accuracy of this predictive modeling from GMD maps, we used a 10-fold cross-validation process. The brain classifier was trained on 90% of the data and tested on the remaining 10% with 10 iterations, so that each participant was used for training the model in nine folds and for testing the accuracy of its prediction in the remaining fold. Ten-fold cross-validation is within the range of typically recommended folds (between 5 and 10) and allowed for a large training sample size at each iteration 45,46. Default regularization parameters were used for all machine-learning analyses to avoid overfitting of the model to the data. We used four metrics to assess the accuracy of the model predictions: the mean squared error (MSE) of prediction, the root mean squared error (RMSE), the mean absolute error (MAE), and the correlation between the model predictions (from the 10 hold-out test samples) and observed log(k)’s (prediction-outcome correlation).
Test of the validity of predicted log(k) in Study 1
To test the reliability of the predictions, we used permutation tests assessing the statistical significance of the accuracy metrics (MSE, RMSE, MAE and prediction-outcome correlation). More precisely, 5000 iterations of randomly permuting the log(k) values were used to generate null distributions of these four metrics and thus to assess the probability of: (MSE < actual MSE), (RMSE < actual RMSE), (Mean abs. error < actual Mean abs. error) and of (prediction-outcome correlation < actual prediction-outcome correlation) under the null hypothesis. To further confirm the validity of out-of-sample predictions of log(k), we performed correlation tests between the predicted log(k) and: (1) calculated log(k) values for the ITC task performed seven weeks later (at the end of the dietary intervention); (2) the urgency trait subscale of the Impulsive Behavior Short Scale–8 (I-8). Since we had directional hypotheses, we used one-tailed correlation tests for all correlations between predicted and observed log(k).
Predictions of the brain pattern in an independent sample of healthy participants in Study 2
To assess the predictions of the brain classifier developed in Study 1 in participants of Study 2, we calculated the dot product between the predictive weight map and the grey matter density map of each participant of Study 2. The dot product (computed as a linear combination of the participant’s voxel grey matter density multiplied by voxel weight across the brain), plus the classifier’s intercept, provides a pattern response and thereby a predicted value of log(k) for each participant. This allowed us to test the correlations between the predicted log(k) values and: (1) the actual log(k) values computed in the sample; (2) the average of positive and negative urgency measures from the UPPS-P Impulsive Behavior Scale.
Predictions of the brain pattern in patients with neurodegenerative dementia in Study 3
To assess the predictions of the brain classifier developed in Study 1 in participants of Study 3, we calculated again the dot product as a measure of pattern response and thereby a predicted value of log(k) for each participant of Study 3. This allowed us to test: (1) the correlation between the predicted log(k) and the actual log(k) values (for both monetary and food rewards) across the whole sample (bvFTD patients and matched controls); (2) whether predicted log(k) values could accurately discriminate between bvFTD patients and controls, using a single-interval test (thresholded for optimal overall accuracy). Further, we explored whether the predicted log(k)’s were related to the severity of inhibition deficit (measured by Hayling error score) and of dysexecutive syndrome (i.e., lower FAB total score) among bvFTD patients.
Bootstrapping and thresholding of the predictive brain pattern obtained in Study 1
We used a bootstrapping analysis to detect the brain regions that were the most robust contributors to predict log(k). Sampling with replacement from the initial sample of Study 1 participants generated 5,000 samples. The LASSO-PCR algorithm yielded a predictive brain pattern (voxel weights across the brain) from the data (paired GMD map – log(k) outcome) in each of these 5,000 samples. For each voxel weight in the whole-brain pattern, the probability of being different from 0 (either above or below 0) could be estimated across the 5,000 samples. Thus, two-tailed, uncorrected p-values were calculated for each voxel across the whole brain and false discovery rate (FDR) correction was used to correct for multiple comparisons. Bootstrapped weights were thresholded at q = 0.05 FDR-corrected across the whole weight map, as well as at p = 0.05 uncorrected for display.
Spatial distribution of weights in the predictive brain pattern obtained in Study 1
To further characterize the spatial distribution of regions predicting log(k) and their link to different functional networks, we investigated the similarity between the predictive brain pattern (resulting from the LASSO-PCR procedure) and term-based meta-analytic images 47 representing functional networks that have been previously hypothesized 48 to contribute to temporal discounting, namely brain areas related to valuation, executive control and memory/prospection. We calculated the spatial correlation coefficients (Pearson’s r) between the brain pattern (map of weights) and each of the meta-analytic maps (thresholded meta-analytic uniformity maps from Neurosynth) corresponding to the following list of terms: “value”, “reward”, “emotion”, “affect”, “executive”, “conflict”, “cognitive control”, “attention”, “planning”, “imagery”, “memory”, “episodic memory”. These spatial correlations provide descriptive insight into the importance of the contribution of GMD within specific functional networks to predict individual differences in delay discounting 14,49.
Development and cross-validation of a structural brain signature predicting delay discounting in healthy adults (Study 1)
Individual differences in impulsivity
On average, participants had a fitted log(k) parameter of -5.94 (median log(k)=-5.49, corresponding to k = 0.0041). Discounting rates were characterized by substantial individual differences (SD = 2.00), with log(k) ranging from − 11.92 to -2.16. These individual differences were very stable over a 7-week period as reported previously 14. On the I-8 subscale of urgency trait, participants’ average scores varied between 1 and 5 (mean = 2.72; median = 2.5; SD = 0.84). Log(k) showed a trend for a weak positive correlation with the urgency trait (R = 0.17, p = 0.06, 95%-CI= [-0.009, 0.35]).
Cross-validated predictions of delay discounting - Validity of predicted log(k) in healthy participants
The 10-fold cross-validation procedure revealed a significant accuracy of the brain-based prediction (see Fig. 2A and 2B and Supplementary Fig. 1): the predictions had a mean squared error of 3.45 (permutation test: p = 0.0026), a root mean squared error of 1.86 (permutation test: p = 0.0026), a mean absolute error for predicted log(k) of 1.46 (permutation test: p = 0.0022), and a cross-validated prediction-outcome correlation of R = 0.35 (permutation test: p = 0.0028) (Fig. 2C).
Further, supporting the reliability and conceptual validity of the brain-predicted log(k)’s, we found that brain-based predictions at baseline significantly correlated with (out-of-sample) log(k)’s computed from the ITC task performed seven weeks later (R = 0.34, p < 0.001, 95%-CI= [0.18, 1]) (Fig. 2D). This suggests that a relatively stable part of the between-person variability in delay discounting was explained by individual differences in brain structure. Moreover, higher brain-predicted log(k) values were associated with higher self-reported urgency (R = 0.20, p = 0.037, 95%-CI= [0.01, 0.37]) (Fig. 2E).
Like the actual measures of log(k) (see 14), brain-based predictions of log(k) did not significantly correlate with age (R=-0.11, p = 0.24, 95%-CI= [-0.29, 0.07]), education (R=-0.15, p = 0.10, 95%-CI= [-0.33, 0.03]), income (R=-0.12, p = 0.21, 95%-CI= [-0.30, 0.07]), BMI (R= -0.04, p = 0.66, 95%-CI= [-0.22, 0.14]), and percentage of body fat (R= -0.13, p = 0.18, 95%-CI= [-0.31, 0.06]) (see more details in Supplementary Fig. 2).
Performance of the Structural Impulsivity Signature in a second independent sample of healthy participants (Study 2)
Study 2 tests the predictions of the Structural Impulsivity Signature (SIS) in a second MRI dataset of healthy participants, that has used a different protocol, scanner, different preprocessing pipeline, in a socio-demographically different participant population.
Individual differences in impulsivity
The mean log(k) parameter in Study 2 was − 4.09 (median log(k)=-3.94, corresponding to a k of 0.019). Individual differences in the discounting parameter were less variable (SD = 0.98) as compared to Study 1, with log(k) ranging from − 7.08 to -2.12. Participants had average urgency trait scores (means of positive and negative urgency) varying between 1.00 and 3.01 (mean = 1.76; median = 1.68; SD = 0.48). In Study 2, log(k) had a trend for a negative correlation with urgency (R=-0.14, p = 0.06, 95%-CI= [-0.29, 0.008]). Therefore, in Study 2, the discounting rate does not seem to be related to individual differences in impulsivity.
Brain-based predictions of impulsivity - Validity of predicted log(k) in a second independent sample of healthy participants
For each participant in Study 2, we calculated the predicted individual log(k) as the dot-product between the weight map developed in Study 1 and the individual GMD map. We then tested whether predicted log(k) correlated with observed individual log(k) and with the impulsivity trait of urgency (UPPS subscales). While we did not find a significant link between predicted and observed log(k) in Study 2 (R = 0.06, p = 0.21, 95%-CI= [-0.07, 1]), predicted log(k) was positively associated with urgency scales (R = 0.15, p = 0.047, 95%-CI= [0.002, 0.30], see Fig. 2F), as in Study 1. Thus, the results of Study 2 partially validate the developed structural brain signature as a brain signature of impulsivity.
Validation of the structural brain signature in a clinical sample of bvFTD patients and matched controls (Study 3)
Our last analysis step aimed at further testing the generalizability of the SIS by evaluating its validity in a patient population that is characterized by impulsivity. Study 3 employed a distinct protocol from Studies 1 and 2 (different ITC task, different MRI scanner and parameters), and in a different, older population including dementia patients with substantial structural atrophy. This further allowed us to investigate the clinical relevance of the SIS (1) for classifying patients with bvFTD differently from matched control participants and (2) for predicting the core symptoms of disinhibition and executive deficits in patients with bvFTD 8 .
Differences of impulsivity between bvFTD patients and healthy controls
In line with the core symptoms of this disorder, bvFTD patients presented significantly higher delay discounting (i.e. more impatient or impulsive choices) compared to controls, for both money rewards and food rewards (see 35). They also showed higher inhibition deficit (Hayling-error score; t = 5.71, p < 0.001, Cohen’s d = 1.60, 95%-CI=[0.87, 2.33]) and lower executive performances (FAB score; t=-7.31, p < 0.001, Cohen’s d=-2.00, 95%-CI=[-1.23, -2.77]) compared to controls (see Supplementary table 1).
Brain-based predictions of impulsivity – Validity of predicted log(k) in bvFTD patients
To investigate the predictive validity of our classifier in Study 3, we first tested whether predicted log(k)’s (obtained from the brain pattern applied to each participant’s grey matter density map) were correlated with actual log(k)’s calculated in this study across the whole sample (patients and controls). This analysis showed that the predicted log(k) values were positively correlated with actual log(k) values, for both monetary rewards (R = 0.30, p = 0.03, 95%-CI= [0.03, 1], mean absolute error of 2.08) and for food rewards (R = 0.40, p = 0.006, 95%-CI= [0.15, 1], mean absolute error of 2.65) (see Fig. 3.A and 3.B).
We next tested whether the SIS predictions could distinguish bvFTD patients from controls. As expected, we found that brain-predicted log(k) was significantly higher in bvFTD patients than in controls (t = 3.60, p = 0.0009, Cohen’s d = 1.09, 95%-CI=[0.41, 1.76] – see Fig. 3.C). Notably, brain-predicted log(k) significantly predicted whether a grey matter density map was from a bvFTD patient or from a control participant, with a classification accuracy of 81% (p = 0.002, sensitivity = 87.5%, specificity = 72.2%, - see Fig. 3.D). Interestingly, the actual log(k)’s calculated for monetary and food rewards in this sample revealed slightly lower predictive accuracies and especially lower specificities: 73.7% accuracy for monetary rewards (p = 0.07, sensitivity = 100%, specificity = 37.5%,) and 76.3% for food rewards (p = 0.01, sensitivity = 100%, specificity = 47.1%).
We next investigated the relationship between brain-predicted log(k) and clinical measures of bvFTD core symptoms of disinhibition and executive deficits. Across both the patient and control groups, higher predicted log(k) was associated with higher inhibition deficit (higher Hayling-error score; R = 0.55, p = 0.0002, 95%-CI= [0.30, 0.74]) and higher executive troubles (lower FAB score; R=-0.56, p = 0.0001, 95%-CI= [-0.74, -0.30]). More interestingly, even within the group of bvFTD patients, higher predicted log(k) was associated with higher inhibition deficit (higher Hayling-error score; R = 0.52, p = 0.01, 95%-CI= [0.14, 0.77]) and higher executive troubles (lower FAB score; R=-0.43, p = 0.04, 95%-CI= [-0.71, -0.03]) (see Fig. 3.E and 3.F). Further, we checked that predicted log(k) was still significantly related to lack of inhibition (i.e., higher Hayling-error scores; B = 8.63, p = 0.02, 95%-CI= [1.51, 15.7]) within bvFTD patients even after controlling for executive function deficit; this added result showed that the relationship between brain-based predictions and disinhibition symptom was not only due to shared variance with the severity of dysexecutive syndrome. Together, these findings show that the SIS significantly and accurately classified bvFTD patients from matched controls, and that it tracked the severity of key symptoms in these patients.
Spatial distribution of weights in the structural brain signature (Study 1)
Thresholded pattern of structural brain signature
Bootstrapping results revealed the positive and negative weights that most strongly contributed to GMD-based prediction of individual differences in delay discounting. At a threshold of q = 0.05 FDR-corrected, we found two clusters in which grey matter density positively contributed to discounting differences (which means that higher grey matter density was associated with higher impatience); these clusters were in the left lateral parietal cortex (supramarginal gyrus) and left lateral occipital cortex (superior division). At a threshold of p = 0.001 uncorrected, we found additional clusters contributing positive weights, especially in regions of the valuation system 50 such as the right orbitofrontal cortex (OFC), ventromedial prefrontal cortex (vmPFC) and right ventral striatum.
At q = 0.05 FDR-corrected, there was one cluster in the posterior cingulate cortex (PCC) and adjacent lingual gyrus (including retrosplenial cortex) in which grey matter density contributed negatively to discounting differences (i.e., in which lower grey matter density was associated with higher impatience). At a threshold of p = 0.001 uncorrected, other important regions contributing negative weights were found in the left hippocampus, the right anterior insulae (AI), dorsal anterior cingulate cortex (ACC), and amygdalae. For display purposes, the bootstrapped weight map is displayed in Fig. 3A at a more comprehensive threshold (p = 0.05 uncorrected, see also Supplementary table 2).
Similarity of structural brain signature to meta-analytic maps
When comparing the predictive map of log(k) with meta-analytic uniformity maps 47, we observed that the highest similarities (spatial correlation r’s > 0.1 in absolute value) were with the “Emotions”, “Affect”, “Conflict” and “Imagery” meta-analytic maps (Fig. 4B). These spatial correlations were all negative, indicating that greater grey matter density in areas related to emotions, affect, conflict processing, and imagery contributes to predicting lower delay discounting or more ‘patient’ decision-making (or conversely, lower grey matter density in these areas predicts higher discounting and more impulsive decision-making). The “Emotions”, “Affect”, “Conflict” and “Imagery” meta-analytic maps correspond to overlapping functional networks (see Fig. 4.B). Among the most overlapping regions between these four networks (in red), the AI and dorsal ACC, corresponding to robust negative weights in the brain pattern, are known to be major hubs of the salience network 51.
Spatial distribution of brain regions contributing to higher predicted log(k) in bvFTD (Study 3)
To identify the main brain regions which contributed to differentiate bvFTD patients from controls on the brain-predicted log(k), we contrasted bvFTD patients versus controls in terms of voxel-wise pattern expression of the predictive map of log(k). To this end, for each bvFTD patient and each control participant, we computed an ‘importance map’ as the unsummed matrix dot product between the predictive structural weight map and the individual grey matter density map. Since higher resulting dot product contributes to higher predicted discounting, the importance map shows which brain regions contributed to increase (or decrease) predicted discounting in each individual. We performed a t-test contrasting bvFTD patients and controls (bvFTD > controls) on the resulting importance maps, with a family-wise error (FWE) correction applied to p-values to correct for multiple comparisons across the brain (see Fig. 5.C). This contrast shows the regions in which structural atrophy contributed positively to higher predicted discounting in bvFTD than in controls (regions in red). These included the OFC, anterior insulae, dorsal ACC, striatum, thalamus, amygdalae, hippocampus, and middle temporal regions. These regions corresponded to areas combining the presence of negative weights in the predictive brain pattern (i.e., voxels for which higher GMD predicts lower discounting and more patient decision-making, shown in Fig. 5.B) and the presence of significant grey matter atrophy due to bvFTD pathology (see atrophy pattern in Fig. 5.A). Thus, the contrast shown in Fig. 5.C also maps the regions in which the SIS is the most similar to bvFTD atrophy pattern.
Impulsive and maladaptive decision-making is a transversal feature of many mental disorders, especially prominent in behavioral-variant frontotemporal dementia (bvFTD). Yet, its relationship with individual brain characteristics, in particular brain structure, is still debated. Here, we used a machine learning technique to develop a brain signature (i.e., a multi-variate brain model) of individual differences in delay discounting—a facet of impulsivity—based on whole-brain grey matter density patterns. We performed out-of-sample cross-validation in a first sample of 117 healthy adults (Study 1) used for brain signature development. We further tested the generalizability of this brain signature developed in Study 1 in two independent studies: a second sample of 166 healthy adults (Study 2) and a clinical study including 24 bvFTD patients and 18 matched controls (Study 3). Individual differences of whole-brain grey matter density reliably predicted individual differences in discounting rates in the first sample of healthy adults but not in the second independent sample. However, the brain signature predicted individual differences of urgency (a subcomponent of impulsivity according to the UPPS model) with small-to-moderate effect sizes in both the first and the second samples of healthy adults. Most importantly, in the clinical study, we found that this structural signature of impulsivity (SIS) separated bvFTD patients from controls with 81% accuracy and that it significantly predicted not only individual differences in delay discounting across participants but also inhibition deficit (objectively assessed from the Hayling test), even within the group of bvFTD patients. Thus, the SIS might be more closely and reliably related to the broader concepts of impulsivity, urgency, and inhibition deficits rather than to specifically delay discounting, which may be more driven by cultural and educational factors than trait urgency. In sum, our results suggest that: 1) it is possible to predict individual differences in impulsivity from whole-brain structure and 2) this novel brain signature is sensitive to the structural atrophy that is characteristic of bvFTD, making it a novel candidate neuromarker for improving bvFTD diagnosis.
The identification of the SIS advances our knowledge of the neurobiology underlying individual differences in impulsivity. Higher discounting (i.e., greater impulsivity) was associated with higher grey matter density in clusters of the lateral parietal and occipital cortex as well as in regions of the OFC, vmPFC, ventral striatum, lateral PFC, precentral gyrus, and precuneus. Functional activation of these regions during intertemporal choices and in response to rewards has previously been shown to predict higher discounting 14,52. The SIS obtained from Study 1 also revealed regions in which greater grey matter density contributes to lower individual impulsivity. Among the strongest negative contributors, we found clusters corresponding to hub regions of the salience network (anterior insulae, dorsal ACC, amygdalae). Dorsal ACC and anterior insula were also consistently found as significant regions predicting delay discounting from whole-brain functional MRI 14. These regions are associated with the processing of emotionally significant internal and external stimuli 51,53,54 and awareness of present and future affective states 55; they are also supposed to be involved in switching between large-scale networks to facilitate access to attention and working memory resources in the presence of a salient event 56. These areas are also known to be involved in cognitive conflict processing 57,58 and previous studies have shown their response to difficult choices (characterized by choice conflicts between options) during delay discounting 59. Thus, our results suggest that more impulsive individuals might be those for whom lower affective, attentional, and conflict processing would lead to more impulsive decision-making, favouring immediately rewarding options over long-term consequences of behavior.
The SIS has the potential to contribute to the early diagnosis of conditions characterized by high impulsivity, such as bvFTD. Brain signatures can in particular help the diagnosis of conditions involving brain lesions that are sometimes difficult to detect by mere visual inspection of MRI scans, especially at early stages of the disease. In addition, brain signatures can constitute neuroimaging markers with diagnostic value that can be used across different samples and populations 11,60. The SIS may contribute to the diagnosis of bvFTD by complementing other brain models able to detect bvFTD. A few previous studies successfully trained structural MRI classifiers for the specific purpose of distinguishing FTD patients from controls (e.g., 61–63). These bvFTD classifiers have shown their accuracy to detect patients with clear structural brain damage but their ability to distinguish individuals at risk of developing FTD due to genetic mutations is likely to be limited to the period just before symptom onset 64,65. Under the hypothesis of a continuum of marked impulsivity in presymptomatic individuals and patients 16, the SIS might serve the early prediction and monitoring of bvFTD before symptom onset. Impulsive behaviors may be present in an attenuated form long before clinical diagnosis and hard to detect with traditional clinical methods. A neuromarker predicting impulsivity may be sensitive to specific brain modifications that appear very early in individuals predisposed to FTD (possibly as neurodevelopmental lesions 66) and would thus allow to enhance the monitoring of clinical signs of these subtle behavioral changes. Future tests of this brain signature in presymptomatic populations will allow to evaluate these potential clinical applications.
As it predicts nearly 30% of the variance of inhibition deficit among bvFTD patients, the SIS may be sensitive to lesions in a structural network underlying the core bvFTD symptom of disinhibition. In addition to its potential contribution to the early detection of presymptomatic individuals, this brain signature may thus aid differential diagnosis and provide insight into the neuropsychological profiles of patients. The SIS may for instance help to distinguish bvFTD from other neurodegenerative or neuropsychiatric conditions with different core symptoms. The differential diagnosis of Alzheimer’s disease and bvFTD can in particular be challenging. Using neuromarkers such as the SIS in cases of diagnostic uncertainty potentially impacting the choice of treatment could therefore be highly valuable 67 and should be an avenue for future studies. Moreover, the SIS could become a useful tool to disentangle the phenotypic heterogeneity within bvFTD population 68. The characterization of different clinical and behavioral profiles within the bvFTD spectrum could help to better understand the pathology, and to better adapt treatments according to patients’ specific needs.
Despite holding promises for future clinical applications, we note that our results also point at challenges in generalizing the brain signature to other independent samples of healthy adults. We were successful at predicting delay discounting from whole-brain grey matter in a first rather homogenous sample of healthy adults (male participants, controlled experimental conditions) showing significant variability in terms of impulsivity and a positive correlation between the discounting rate and urgency. In a second independent sample of healthy adults with lower variance of impulsivity and a slightly negative correlation between the discounting rate and urgency, we could not replicate the association with measured discounting rates but found evidence of the conceptual validity (i.e., a link with the urgency trait) of brain-based predictions. This suggests that the variance captured by the SIS developed in the first sample is more reliably related to individual differences in urgency than to individual differences in discounting. The fact that urgency was slightly negatively correlated with the discounting rate in the second healthy sample questions the idea that delay discounting necessarily captures individual differences in impulsivity. These two constructs overlap but are not equivalent and previous studies have already reported an absence of link between delay discounting and some psychometric measures of impulsivity (e.g., 69,70). Discounting rate is also a state-dependent variable 71 and depends on situational factors such as cultural and social context 72. In addition, the links between personality and discounting rates may depend on participants’ cognitive abilities 73. Therefore, association between delay discounting and other measures of trait impulsivity may vary according to samples and studies. A promising approach for future studies would therefore be to predict latent variables that underlie different observed variables related to the same concept of impulsivity (instead of only one observed variable such as the discounting rate), which might achieve better performance in terms of replicability and generalizability 9. Although multivariate brain signatures can be replicable with moderate sample sizes 74, future studies aiming to develop brain signatures of impulsivity could also benefit from using larger and more diverse samples 75. More generally, we note that our results suggest a relatively small contribution of interindividual variability in brain structure to interindividual variability in impulsivity among healthy adults. Effect sizes of associations between predicted and observed impulsivity are however in line with those reported for most brain signatures of behavioral individual differences using structural features 9. Moreover, like variability in brain structure, variability in genotype accounts for a rather small part of the variance of impulsivity 76. The magnitude of associations between brain structure and behaviors may be limited in the general population but these associations might be more salient within populations with a marked variability of both brain and behavior such as patients with neurodegenerative conditions.
In conclusion, our results advance our knowledge of the association between impulsivity and brain structure in healthy adults and in patients with bvFTD. They also point at inherent challenges in developing replicable and generalizable brain signatures of individual differences based on brain structure. By identifying a structural network associated with individual differences in discounting rates, our results provide insight into the potential neurobiological bases of trait impulsivity (and in particular its urgency component). The good performance of the SIS among patients with bvFTD suggests a possible continuum of brain-impulsivity relationship across healthy and clinical conditions. Most noteworthy, the SIS separates bvFTD patients from controls with high accuracy, pointing at the potential clinical value for the diagnosis of bvFTD, in particular for the purpose of stratifying this heterogenous condition. MRI can be instrumental to confirm an FTD diagnosis 67 and the SIS only requires a preprocessed T1-weighted scan to reach a prediction. It holds promise as a phenotypic marker in patients with neurodegenerative or psychiatric conditions associated with high impulsivity. Future studies could test its clinical potential and whether this brain signature could be used in a real-life patient workflow.
Acknowledgments
This study was funded by an ANR Tremplin-ERC grant to HP, a Sorbonne Emergence Grant to HP and LK, and an ERC Starting Grant to LK. Study 2 was funded by National Cancer Institute Grants R01-CA-170297 to J.W.K. and C.L. and R35-CA-197461 to C.L. Study 3 was funded by grant ANR-10-IAIHU-06 from the program ‘Investissements d’avenir’, by grant FRM DEQ20150331725 from the foundation ‘Fondation pour la recherche médicale’, and by the ENEDIS company. FB was supported by Fundação para a Ciência e Tecnologia (CEECIND/03661/2017). We thank all the participants and organizers of the three studies mentioned in this paper as well as all the students who contributed to help data collection.
- Chamberlain SR, Sahakian BJ (2007) The neuropsychiatry of impulsivity. Curr Opin Psychiatry 20:255–261
- Chamberlain SR, Stochl J, Redden SA, Grant JE (2018) Latent traits of impulsivity and compulsivity: toward dimensional psychiatry. Psychol Med 48:810–821
- Goodwin BC, Browne M, Hing N, Russell AM (2017) Applying a revised two-factor model of impulsivity to predict health behaviour and well-being. Personal Individ Differ 111:250–255
- Rogers MM, Kelley K, McKinney C (2021) Trait impulsivity and health risk behaviors: A latent profile analysis. Personal Individ Differ 171:110511
- Bjork JM, Momenan R, Hommer DW (2009) Delay discounting correlates with proportional lateral frontal cortex volumes. Biol Psychiatry 65:710–713
- MacNiven KH, Leong JK, Knutson B (2020) Medial forebrain bundle structure is linked to human impulsivity. Sci Adv 6:eaba4788
- Pehlivanova M et al (2018) Diminished cortical thickness is associated with impulsive choice in adolescence. J Neurosci 38:2471–2481
- Rascovsky K et al (2011) Sensitivity of revised diagnostic criteria for the behavioural variant of frontotemporal dementia. Brain 134:2456–2477
- Genon S, Eickhoff SB, Kharabian S (2022) Linking interindividual variability in brain structure to behaviour. Nat Rev Neurosci 23:307–318
- Kragel PA, Koban L, Barrett LF, Wager TD (2018) Representation, Pattern Information, and Brain Signatures: From Neurons to Neuroimaging. Neuron 99:257–273
- Woo C-W, Chang LJ, Lindquist MA, Wager T (2017) D. Building better biomarkers: brain models in translational neuroimaging. Nat Neurosci 20:365–377
- Anokhin AP, Grant JD, Mulligan RC, Heath AC (2015) The Genetics of Impulsivity: Evidence for the Heritability of Delay Discounting. Biol Psychiatry 77:887–894
- Kirby KN (2009) One-year temporal stability of delay-discount rates. Psychon Bull Rev 16:457–462
- Koban L et al (2023) An fMRI-based brain marker of individual differences in delay discounting. J Neurosci
- Amlung M et al (2019) Delay Discounting as a Transdiagnostic Process in Psychiatric Disorders: A Meta-analysis. JAMA Psychiatry 76:1176
- Godefroy V et al (2023) Altered delay discounting in neurodegeneration: insight into the underlying mechanisms and perspectives for clinical applications. Neurosci Biobehav Rev 105048
- Ballard K, Knutson B (2009) Dissociable neural representations of future reward magnitude and delay during temporal discounting. NeuroImage 45:143–150
- Bergström F, Lerman C, Kable JW (2024) Reduced cortical complexity in ventromedial prefrontal cortex is associated with a greater preference for risky and immediate rewards. Imaging Neurosci 2:1–14
- Cooper N, Kable JW, Kim BK, Zauberman G (2013) Brain activity in valuation regions while thinking about the future predicts individual discount rates. J Neurosci 33:13150–13156
- Hare TA, Hakimi S, Rangel A (2014) Activity in dlPFC and its effective connectivity to vmPFC are associated with temporal discounting. Front Neurosci 8:50
- Lebreton M et al (2013) A critical role for the hippocampus in the valuation of imagined outcomes. PLoS Biol 11:e1001684
- Li N et al (2013) Resting-State Functional Connectivity Predicts Impulsivity in Economic Decision-Making. J Neurosci 33:4886–4895
- van den Bos W, Rodriguez CA, Schweitzer JB, McClure SM (2014) Connectivity strength of dissociable striatal tracts predict individual differences in temporal discounting. J Neurosci 34:10298–10310
- Bartra O, McGuire JT, Kable JW (2013) The valuation system: a coordinate-based meta-analysis of BOLD fMRI experiments examining neural correlates of subjective value. NeuroImage 76:412–427
- Levy DJ, Glimcher PW (2012) The root of all value: a neural common currency for choice. Curr Opin Neurobiol 22:1027–1038
- Rangel A, Clithero JA (2012) Value normalization in decision making: theory and evidence. Curr Opin Neurobiol 22:970–981
- Chare L et al (2014) New criteria for frontotemporal dementia syndromes: clinical and pathological diagnostic implications. J Neurol Neurosurg Psychiatry 85:865–870
- Karageorgiou E, Miller B (2014) Frontotemporal Lobar Degeneration: A Clinical Approach. Semin Neurol 34:189–201
- Beagle AJ et al (2020) Amount and delay insensitivity during intertemporal choice in three neurodegenerative diseases reflects dorsomedial prefrontal atrophy. Cortex J Devoted Study Nerv Syst Behav 124:54–65
- Bertoux M, O’Callaghan C, Flanagan E, Hodges JR, Hornberger M (2015) Fronto-striatal atrophy in behavioral variant frontotemporal dementia and Alzheimer’s disease. Front Neurol 6:147
- Chiong W et al (2016) Neuroeconomic dissociation of semantic dementia and behavioural variant frontotemporal dementia. Brain J Neurol 139:578–587
- Manuel AL et al (2020) Interactions between decision-making and emotion in behavioral-variant frontotemporal dementia and Alzheimer’s disease. Soc Cogn Affect Neurosci 15:681–694
- Han D, Qu L, Sun L, Sun Y (2021) Variable selection for a mark-specific additive hazards model using the adaptive LASSO. Stat Methods Med Res 30:2017–2031
- Tibshirani R (1996) Regression shrinkage and selection via the lasso. J R Stat Soc Ser B Methodol 58:267–288
- Godefroy V et al (2024) High delay discounting relates to core symptoms and to pulvinar atrophy in frontotemporal dementia. Preprint https://doi. org/10.1101/2024.05.16.594526
- Kable JW et al (2017) No Effect of Commercial Cognitive Training on Brain Activity, Choice Behavior, or Cognitive Performance. J Neurosci 37:7390–7402
- Groskurth K, Nießen D, Rammstedt B, Lechner CM (2022) The impulsive behavior short scale–8 (I-8): A comprehensive validation of the English-language adaptation. PLoS ONE 17:e0273801
- Whiteside SP, Lynam DR (2001) The five factor model and impulsivity: Using a structural model of personality to understand impulsivity. Personal Individ Differ 30:669–689
- Burgess PW, Shallice T (1997) The hayling and brixton tests
- Dubois B, Slachevsky A, Litvan I, Pillon B (2000) The FAB: a frontal assessment battery at bedside. Neurology 55:1621–1626
- Flanagan EC et al (2016) False recognition in behavioral variant frontotemporal dementia and Alzheimer’s disease—disinhibition or amnesia? Front Aging Neurosci 8:177
- Wileyto EP, Audrain-McGovern J, Epstein LH, Lerman C (2004) Using logistic regression to estimate delay-discounting functions. Behav Res Methods Instrum Comput 36:41–51
- Wager TD, Atlas LY, Leotti LA, Rilling JK (2011) Predicting Individual Differences in Placebo Analgesia: Contributions of Brain Activity during Anticipation and Pain Experience. J Neurosci 31:439–452
- Wager TD et al (2013) An fMRI-Based Neurologic Signature of Physical Pain. N Engl J Med 368:1388–1397
- Poldrack RA, Huckins G, Varoquaux G (2020) Establishment of best practices for evidence for prediction: a review. JAMA Psychiatry 77:534–540
- Scheinost D et al (2019) Ten simple rules for predictive modeling of individual differences in neuroimaging. NeuroImage 193:35–45
- Yarkoni T, Poldrack RA, Nichols TE, Van Essen DC, Wager T (2011) D. Large-scale automated synthesis of human functional neuroimaging data. Nat Methods 8:665–670
- Peters J, Büchel C (2011) The neural mechanisms of inter-temporal decision-making: understanding variability. Trends Cogn Sci 15:227–239
- Koban L, Jepma M, López-Solà M, Wager TD (2019) Different brain networks mediate the effects of social and conditioned expectations on pain. Nat Commun 10:1–13
- Lempert KM, Steinglass JE, Pinto A, Kable JW, Simpson HB (2019) Can delay discounting deliver on the promise of RDoC? Psychol Med 49:190–199
- Menon V (2015) Salience Network. in Brain Mapping 597–611Elsevier, 10.1016/B978-0-12-397025-1.00052-X
- Hariri AR et al (2006) Preference for immediate over delayed rewards is associated with magnitude of ventral striatal activity. J Neurosci 26:13213–13217
- Seeley WW (2019) The salience network: a neural system for perceiving and responding to homeostatic demands. J Neurosci 39:9878–9882
- Seeley WW, Zhou J, Kim E-J (2012) Frontotemporal dementia: what can the behavioral variant teach us about human brain organization? Neuroscientist 18:373–385
- Craig AD (2009) How do you feel — now? The anterior insula and human awareness. Nat Rev Neurosci 10:59–70
- Menon V, Uddin LQ (2010) Saliency, switching, attention and control: a network model of insula function. Brain Struct Funct 214:655–667
- Botvinick MM, Braver TS, Barch DM, Carter CS, Cohen JD (2001) Conflict monitoring and cognitive control. Psychol Rev 108:624
- Van Veen V, Cohen JD, Botvinick MM, Stenger VA, Carter CS (2001) Anterior cingulate cortex, conflict monitoring, and levels of processing. NeuroImage 14:1302–1308
- Hoffman WF et al (2008) Cortical activation during delay discounting in abstinent methamphetamine dependent individuals. Psychopharmacology 201:183–193
- Haller S, Lovblad K-O, Giannakopoulos P, Van De Ville D (2014) Multivariate Pattern Recognition for Diagnosis and Prognosis in Clinical Neuroimaging: State of the Art, Current Challenges and Future Trends. Brain Topogr 27:329–337
- Davatzikos C, Resnick SM, Wu X, Parmpi P, Clark CM (2008) Individual patient diagnosis of AD and FTD via high-dimensional pattern classification of MRI. NeuroImage 41, 1220–1227
- Gonzalez-Gomez R, Ibañez A, Moguilner S (2022) Multiclass characterization of frontotemporal dementia variants via multimodal brain network computational inference. Netw Neurosci 1–29
- Moguilner S et al (2023) Visual deep learning of unprocessed neuroimaging characterises dementia subtypes and generalises across non-stereotypic samples. EBioMedicine 90
- Feis RA et al (2018) Single-subject classification of presymptomatic frontotemporal dementia mutation carriers using multimodal MRI. NeuroImage Clin 20:188–196
- Feis RA et al (2019) A multimodal MRI-based classification signature emerges just prior to symptom onset in frontotemporal dementia mutation carriers. J Neurol Neurosurg Psychiatry 90:1207–1214
- Lee SE et al (2017) Network degeneration and dysfunction in presymptomatic C9ORF72 expansion carriers. NeuroImage Clin 14:286–297
- Sheikh-Bahaei N, Sajjadi SA, Pierce AL (2017) Current Role for Biomarkers in Clinical Diagnosis of Alzheimer Disease and Frontotemporal Dementia. Curr Treat Options Neurol 19:46
- O’Connor CM et al (2017) Behavioral-variant frontotemporal dementia: Distinct phenotypes with unique functional profiles. Neurology 89:570–577
- Lane SD, Cherek DR, Rhoades HM, Pietras CJ, Tcheremissine OV (2003) Relationships Among Laboratory and Psychometric Measures of Impulsivity: Implications in Substance Abuse and Dependence. Addict Disord Their Treat 2:33
- Reynolds B, Ortengren A, Richards JB, de Wit H (2006) Dimensions of impulsive behavior: Personality and behavioral measures. Personal Individ Differ 40:305–315
- Odum AL, Baumann AAL (2010) Delay discounting: State and trait variable. Impulsivity: The behavioral and neurological science of discounting 39–65. American Psychological Association, Washington, DC, US. doi:10.1037/12069-002.
- Schwenke D, Wehner P, Scherbaum S (2022) Effects of individual and dyadic decision-making and normative reference on delay discounting decisions. Cogn Res Princ Implic 7:71
- Hirsh JB, Morisano D, Peterson JB (2008) Delay discounting: Interactions between personality and cognitive ability. J Res Personal 42:1646–1650
- Spisak T, Bingel U, Wager T (2022) Replicable multivariate BWAS with moderate sample sizes. bioRxiv 2022–06
- Marek S et al (2022) Reproducible brain-wide association studies require thousands of individuals. Nature 603:654–660
- Sanchez-Roige S et al (2018) Genome-wide association study of delay discounting in 23,217 adult research participants of European ancestry. Nat Neurosci 21:16–18
There is NO Competing Interest.