Foundation models for data-scarce neuropathic pain research

Neuropathic pain is difficult to treat, and data scarcity makes it challenging to train high-capacity models for drug-response prediction. Our work studies how foundation models can transfer knowledge from larger, pain-agnostic datasets into small, pain-specific neuroimaging cohorts.

The resulting framework combines multimodal information within the target cohort with representations learned from external fMRI data. The paper is available on arXiv.