4D fMRI CrossFormer Enhances AI Screening for Brain Disorders

A research team at InfoLab, Sungkyunkwan University (SKKU) has announced the development of a novel vision‑transformer architecture, the 4D fMRI CrossFormer (4DfCF), designed specifically for the analysis of four‑dimensional functional magnetic resonance imaging (fMRI) data. Led by Professor Tamer Abuhmed, the team’s work builds on recent advances in transformer models to address the unique challenges posed by high‑dimensional neuroimaging datasets, which combine spatial and temporal information across brain scans.

The 4DfCF architecture extends conventional transformer designs by incorporating cross‑modal attention mechanisms that simultaneously process spatial and temporal features of fMRI volumes. This allows the model to capture dynamic neural activity patterns more effectively than existing methods. Early evaluations on benchmark fMRI datasets demonstrate improved accuracy in identifying task‑evoked brain regions and predicting cognitive states, suggesting that the architecture could enhance both basic neuroscience research and clinical diagnostics.

If the 4DfCF model continues to perform well in larger studies, it may become a valuable tool for researchers investigating brain function and for clinicians seeking more precise biomarkers in neurological and psychiatric conditions. The team plans to release the code and pretrained models to the scientific community, aiming to accelerate further developments in neuroimaging analysis.