Researchers at Mass General Brigham have developed BrainIAC, a powerful AI foundation model trained on nearly 49,000 brain MRI scans. Using self-supervised learning, it extracts hidden patterns to estimate brain age, predict dementia risk, detect tumor-related genetic mutations, and forecast survival in brain cancer patients offering earlier warnings and supporting personalized care. Published in Nature Neuroscience and released open-source, BrainIAC outperforms specialized models even with limited labeled data, paving the way for broader AI adoption in brain health.
Glimpse:
BrainIAC, or Brain Imaging Adaptive Core, is a generalizable AI foundation model that analyzes routine brain MRIs to uncover clinically relevant signals. It predicts dementia risk through mild cognitive impairment classification, assesses accelerated brain ageing, identifies key genetic mutations in brain tumors, and estimates survival outcomes for cancer patients. Trained via self-supervised learning on 48,965 diverse scans, it adapts effectively to downstream tasks with minimal fine-tuning data, generalizes across healthy and abnormal images, and excels in few-shot scenarios. This breakthrough enables earlier interventions for neurodegenerative diseases and more precise brain cancer management, with the open-source release empowering global research and clinical adaptation.
A new AI foundation model from Mass General Brigham is set to transform how clinicians interpret brain MRIs, delivering earlier warnings for dementia and improved insights into brain cancer. Named Brain Imaging Adaptive Core (BrainIAC), the tool was developed by researchers including Benjamin Kann, MD, from the Artificial Intelligence in Medicine (AIM) Program. The study, published in Nature Neuroscience on February 5, 2026, demonstrates its ability to analyze routine brain scans for multiple disease-related signals.
Trained on nearly 49,000 diverse brain MRIs using self-supervised learning (contrastive SSL), BrainIAC learns generalized representations of brain structure without needing extensive labeled data. This pre-training allows it to adapt to various tasks, from straightforward MRI sequence classification to complex challenges like detecting isocitrate dehydrogenase (IDH) mutations in tumors or predicting time-to-stroke.
For dementia, BrainIAC identifies mild cognitive impairment (MCI) versus healthy controls, flagging elevated risk years before full symptoms appear. It also estimates “brain age” to highlight accelerated ageing linked to neurodegenerative decline. In brain cancer, it detects tumor-related genetic mutations and forecasts patient survival, aiding personalized treatment decisions and potentially improving outcomes.
Unlike narrowly trained AI systems, BrainIAC performs robustly even with limited labeled data outperforming benchmarks in few-shot (K=1 or K=5) scenarios and maintaining accuracy across artifacts or variations in imaging. Its open-source availability means institutions can fine-tune it locally without requiring thousands of annotated scans, accelerating adoption in research and practice.
The model’s versatility addresses a key gap in brain imaging AI: most existing tools are task-specific and struggle with real-world variability. By leveraging large-scale unlabeled data, BrainIAC unlocks underutilized information in routine MRIs, supporting earlier detection, better prognosis, and tailored interventions for neurological conditions.
This advancement aligns with growing momentum in AI-driven neuroimaging, promising to reduce diagnostic delays, enhance equity in brain health assessments, and inform future biomarker discovery and therapy development.
“BrainIAC is an AI foundation model that is trained on tens of thousands of brain MRI scans to understand how the brain is structured. Using this core baseline knowledge, the tool can then be adapted to identify various brain diseases, determine their severity, and predict future risks from these diseases.”
By
HB Team
