Add dataset card / README
Browse files
README.md
ADDED
|
@@ -0,0 +1,62 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: other
|
| 3 |
+
tags:
|
| 4 |
+
- alzheimer
|
| 5 |
+
- oasis-1
|
| 6 |
+
- structural-mri
|
| 7 |
+
- multimodal
|
| 8 |
+
- dementia-staging
|
| 9 |
+
pretty_name: TriFuse-AD OASIS-1 Three-Stage Dementia Staging
|
| 10 |
+
---
|
| 11 |
+
|
| 12 |
+
# TriFuse-AD: Honest Multimodal Benchmark for Three-Stage Dementia Staging on OASIS-1
|
| 13 |
+
|
| 14 |
+
Code, processed data, results, and paper for a leakage-free benchmark of three-stage
|
| 15 |
+
cognitive classification (CN / VMD / AD) on the OASIS-1 cross-sectional cohort, plus
|
| 16 |
+
the proposed **TriFuse-AD** model (tri-planar CNN + slice-plane Transformer + gated
|
| 17 |
+
demographic fusion).
|
| 18 |
+
|
| 19 |
+
## Key result (honest / negative)
|
| 20 |
+
|
| 21 |
+
On an age-restricted cohort (≥60, 198 subjects) with subject-level repeated 5-fold CV
|
| 22 |
+
(3 seeds, 15 runs/model), **no MRI-only network beats a plain tabular XGBoost
|
| 23 |
+
(Macro-F1 0.474)**, and TriFuse-AD (0.488 ± 0.066) does **not** significantly beat a
|
| 24 |
+
trivial DenseNet late-concat baseline (0.497 ± 0.062; paired permutation p = 0.55). A
|
| 25 |
+
no-MRI structured model reaches Macro-F1 0.480 — most recoverable signal is
|
| 26 |
+
morphometric/demographic, not learned from raw voxels. No clinical / diagnostic /
|
| 27 |
+
SOTA / MCI / cross-site claims.
|
| 28 |
+
|
| 29 |
+
## Repository layout
|
| 30 |
+
|
| 31 |
+
| Path | Contents |
|
| 32 |
+
|------|----------|
|
| 33 |
+
| `src/` | `trifuse` package: data, models, training, eval, analysis |
|
| 34 |
+
| `scripts/` | experiment runner, table/figure/interpretability builders |
|
| 35 |
+
| `configs/` | model configs |
|
| 36 |
+
| `results/` | per-model OOF preds, summaries, tables, 27 figures |
|
| 37 |
+
| `paper/trifuse_ad.md` | full paper draft |
|
| 38 |
+
| `data_processed.zip` | preprocessed 2.5D + 3D arrays + `subjects_clean.csv` (1.5 GB) |
|
| 39 |
+
| `data/raw/*.tar.gz` | OASIS-1 cross-sectional discs 1–12 (16 GB) |
|
| 40 |
+
|
| 41 |
+
## Reproducing
|
| 42 |
+
|
| 43 |
+
```bash
|
| 44 |
+
pip install -r requirements # torch cu128, timm, monai, nibabel, xgboost, sklearn, ...
|
| 45 |
+
unzip data_processed.zip # -> data/processed_2d, processed_3d, metadata
|
| 46 |
+
python scripts/run_experiments.py --grid main # 11 models x 15 runs
|
| 47 |
+
python scripts/run_experiments.py --grid ablation # 6 variants
|
| 48 |
+
python scripts/make_tables.py && python scripts/make_figures.py
|
| 49 |
+
```
|
| 50 |
+
|
| 51 |
+
## Cohort
|
| 52 |
+
|
| 53 |
+
OASIS-1, age≥60 → 198 subjects (CN=98, VMD=70, AD=30). Labels from CDR (0→CN,
|
| 54 |
+
0.5→VMD, ≥1→AD). CDR and MMSE are **never** model inputs (label leakage). One volume
|
| 55 |
+
per subject (`*_111_t88_masked_gfc`).
|
| 56 |
+
|
| 57 |
+
## License / data use
|
| 58 |
+
|
| 59 |
+
The `data/raw/` tarballs are the original **OASIS-1** cross-sectional release
|
| 60 |
+
(Marcus et al., 2007), redistributed here for reproducibility. OASIS data are subject
|
| 61 |
+
to the OASIS data-use terms; if you use them, cite the OASIS project and comply with
|
| 62 |
+
their agreement. Code and derived results in this repo are provided for research use.
|