BrainNorm: A Foundation Model that knows Normal via Semantic Atlas Pretraining
Ranking
Overall
81
Content
100
Popularity
37
Observed public metrics from 1 member.
Merged summary
TL;DR - BrainNorm is a foundation model pretrained on roughly 66,000 structural brain MRI scans to model healthy, age-dependent anatomy. Its localized deviation scores generalize across aging and disease tasks and may help identify clinically meaningful neurodegeneration patterns.
- Contrastive pretraining on healthy cohorts creates a Semantic Atlas Latent space with embeddings for individual brain parcels.
- The model learns parcel-specific healthy-aging trajectories, enabling age-matched comparisons and localized abnormality scoring.
- Evaluations span six cohorts and 25 settings, including zero-shot, few-shot, and linear-probe age and disease prediction.
- Frozen BrainNorm embeddings with linear probing outperformed nine end-to-end-finetuned baselines across the reported classification tasks.
Sources (1)
BrainNorm: A Foundation Model that knows Normal via Semantic Atlas Pretraining
Public signals
Semantic Scholar citations 0 · Semantic Scholar influential citations 0
TL;DR - BrainNorm is a foundation model pretrained on roughly 66,000 structural brain MRI scans to model healthy, age-dependent anatomy. Its localized deviation scores generalize across aging and disease tasks and may help identify clinically meaningful neurodegeneration patterns.
- Contrastive pretraining on healthy cohorts creates a Semantic Atlas Latent space with embeddings for individual brain parcels.
- The model learns parcel-specific healthy-aging trajectories, enabling age-matched comparisons and localized abnormality scoring.
- Evaluations span six cohorts and 25 settings, including zero-shot, few-shot, and linear-probe age and disease prediction.
- Frozen BrainNorm embeddings with linear probing outperformed nine end-to-end-finetuned baselines across the reported classification tasks.