{
  "id": 11333887,
  "title": "AREION: A Foundation Model for Atlas-Agnostic Representation Learning of Brain Connectome",
  "url": "https://urgent.news/2026/10/01/areion-a-foundation-model-for-atlas-agnostic-representation-learning",
  "topic": "science",
  "section": "Science",
  "published": "2026-10-01T00:00:00.000Z",
  "source": {
    "name": "bioRxiv",
    "slug": "biorxiv",
    "url": "https://www.biorxiv.org/content/10.64898/2026.09.26.754551v1?rss=1"
  },
  "original_language": "en",
  "account": "Recent progress in brain foundation models has enabled the creation of generalizable representations from functional magnetic resonance imaging (fMRI) data. However, these models often rely on specific brain atlases, limiting their use in neuroimaging studies represented using different atlases. To overcome this limitation, researchers have introduced AREION, a novel method for atlas-agnostic representation learning of brain connectomes.\n\nAREION achieves atlas independence through two key innovations. The first is the Brain Region Contextualizing technique, which maps brain regions and connectivity entries from various atlases into a single representation space. This is based on the shared spatial, functional, and morphological features of these regions. The second innovation is the Hierarchical Brain Transformer Encoder, which processes region-specific connectivity profiles as variable-length token sequences. This encoder jointly captures both local functional structure and global brain functional organization, all while being conditioned on brain regional context. Importantly, this encoder does not require a fixed number of regions, making it versatile.\n\nBy combining these two techniques, AREION enables the learning of generalizable representations across brain atlases with different parcellation resolutions and connectivity dimensions. The model is pretrained on a large-scale multi-atlas cohort of fMRI data using self-supervised modeling. It is then evaluated across a range of downstream tasks. The results show that AREION outperforms strong supervised baselines and other fMRI foundation models, even though it employs fewer parameters. This demonstrates the model's strong transferability to unseen atlases, its ability to generalize to external datasets, and its excellent interpretability at the brain regional level.",
  "summary": "Brain foundation models have shown promising advances in learning generalizable representations from functional magnetic resonance imaging (fMRI) data. However, existing models are typically tied to specific brain atlases and cannot be applied to fMRI data represented using different atlases without full retraining, thereby severely limiting their broader utility across neuroimaging studies. To…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}