Add model card
Browse files
README.md
ADDED
|
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: cc-by-nc-4.0
|
| 3 |
+
tags:
|
| 4 |
+
- brain-age
|
| 5 |
+
- neuroimaging
|
| 6 |
+
- mri
|
| 7 |
+
- 3d-cnn
|
| 8 |
+
- pytorch
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
+
# BrainAge — Dual-Branch Brain Age Predictor (resnet18)
|
| 12 |
+
|
| 13 |
+
3D RESNET18 + tabular MLP trained on **6,050 healthy T1w brain MRIs** (0–86 years).
|
| 14 |
+
|
| 15 |
+
## Checkpoints
|
| 16 |
+
|
| 17 |
+
| File | Description |
|
| 18 |
+
|------|-------------|
|
| 19 |
+
| `brainage_resnet18.pt` | Phase A: lifespan pretraining (all ages) |
|
| 20 |
+
| `brainage_resnet18_finetune.pt` | Phase B: fine-tuned for pediatric (0–25 y) |
|
| 21 |
+
|
| 22 |
+
## Training data
|
| 23 |
+
|
| 24 |
+
[bilalEthizo/BrainAge-Golden-Preprocessed](https://huggingface.co/datasets/bilalEthizo/BrainAge-Golden-Preprocessed)
|
| 25 |
+
|
| 26 |
+
## Architecture
|
| 27 |
+
|
| 28 |
+
- Image: resnet18 3D CNN, input 128×144×112
|
| 29 |
+
- Tabular: MLP over 86-dim vector (70 regions + sex + site)
|
| 30 |
+
- Fusion: concat → 64 → 1 (regression in years)
|
| 31 |
+
- Loss: HuberLoss(delta=3.0)
|
| 32 |
+
- Training: fp16 mixed precision, cosine LR, age-bin weighted sampler
|