Two-Step Task Reinforcement Learning fMRI (ds007474)

This release holds the imaging data of a Caltech and UCLA study by John O'Doherty's group on how people differ in the way they learn from rewards. Healthy adults played a variant of the two-step task, a sequential choice game that separates model-based planning from model-free habit-like learning, while their brains were scanned with fMRI. The imaging data sit on OpenNeuro as ds007474 under CC0. Event timings and behavioral responses are on OSF (doi:10.17605/OSF.IO/CTFZD), and task and analysis code on GitHub. The data suit work on reward learning, value signals in the striatum and prefrontal cortex, and brain-behavior links across individuals.

Composition

189 adults were scanned. Every subject has a T1-weighted and a T2-weighted structural image, two task fMRI runs with single-band reference images, and a pair of opposite phase-encoding field maps per run. One subject lacks the field maps of the first run. Participants were recruited in greater Los Angeles at ages 18 to 65 and screened by a psychiatrist; people reporting substance use, anxiety or depressive disorders, or psychiatric medication, were not included. The two-step task was paired with a second task on habits in the same session, and the subject id records whether the two-step task came first or second. Five subjects saw only state-contingent rewards. The paper analysed 179 of the 189 after excluding 10 for head motion; of those 179, 105 were female and the mean age was 30.4 years. Sex and age per subject are not part of the OpenNeuro release.

Acquisition

All scans come from one Siemens Prisma 3T with a 32-channel head coil at the Caltech Brain Imaging Center. Functional runs use multiband EPI (factor 4) with 2 mm isotropic voxels and TR 1.12 s. T1w and T2w images are 0.9 mm isotropic. Data are organized in BIDS.

Annotations

There are no image labels. The derivatives folder holds group-level SPM t-maps for model-based and model-free value and prediction error regressors.

Known limitations

• Single site and scanner, and a screened healthy sample, so results may not transfer to clinical populations.
• No participants.tsv, so demographics cannot be matched to subjects.
• Behavioral data needed to model the task are hosted separately on OSF.