🧠 BrainAnytime Demo
BrainAnytime: Anatomy-Aware Cross-Modal Pretraining for Brain Image Analysis
This demo showcases the BrainAnytime model for multi-modal 3D brain image analysis. Select a task and modality combination, then run inference to see predictions and attention visualization.
Select a task and modality combination to run inference on pre-selected samples. Each sample is from the ADNI training set and has been verified for accuracy.
Configuration
区分认知正常 (CN) 与阿尔茨海默病 (AD)
T: Only T1-weighted MRI
Sample Preview & Attention Visualization
Prediction Result
Click 'Run Inference' to see results
About BrainAnytime
Paper: BrainAnytime: Anatomy-Aware Cross-Modal Pretraining for Brain Image Analysis with Arbitrary Modality Availability
Conference: MICCAI 2026 (Early Accept, Top 9%)
Key Features
- Multi-modal Support: T1, T2, Flair, PET
- Missing Modality Robustness: Handles arbitrary missing modality combinations
- Anatomy-Aware: Uses AAL116 brain atlas for adaptive masking
- Attention Visualization: See which brain regions the model focuses on
Supported Tasks
| Task | Type | Description |
|---|---|---|
| CN vs AD | Classification | Distinguish Normal vs Alzheimer's |
| CN vs MCI | Classification | Distinguish Normal vs Mild Cognitive Impairment |
| MMSE | Regression | Predict cognitive score (10-30) |
| AGE | Regression | Predict age from brain MRI |
Links
- GitHub: https://github.com/guangqianyang/BrainAnytime
- Model Weights: https://huggingface.co/Simmonstt/BrainAnytime
- Paper: arXiv:2605.13059
Citation
@misc{yang2026brainanytime,
title={BrainAnytime: Anatomy-Aware Cross-Modal Pretraining for Brain Image Analysis},
author={Yang, Guangqian and Ding, Tong and others},
year={2026},
eprint={2605.13059},
archivePrefix={arXiv},
}