epics-lab/BrainFMBench · auto generated from models/*/results.json
Held out test, mean ± std over 5 seeds. Click a dataset column to sort by that task; click again to reverse. There is no single overall score.
| Model ⇅ | Sex classification (balanced accuracy, higher is better) | Age prediction (MAE in years, lower is better) | BMI prediction (MAE, lower is better) | |||
|---|---|---|---|---|---|---|
| HBN ⇅ | NKI ⇅ | HBN ⇅ | NKI ⇅ | HBN ⇅ | NKI ⇅ | |
| 3D-Neuro-SimCLR | 0.768 ± 0.049 | 0.835 ± 0.023 | 1.64 ± 0.12 | 5.86 ± 0.58 | 3.02 ± 0.29 | 3.52 ± 0.17 |
| AnatCL (Global) | 0.723 ± 0.060 | 0.738 ± 0.031 | 1.78 ± 0.18 | 6.26 ± 0.43 | 3.39 ± 0.33 | 3.92 ± 0.22 |
| AnatCL (Local) | 0.707 ± 0.063 | 0.765 ± 0.024 | 1.85 ± 0.10 | 6.19 ± 0.47 | 3.34 ± 0.40 | 3.92 ± 0.25 |
| Brain2Vec | 0.727 ± 0.048 | 0.856 ± 0.044 | 1.49 ± 0.09 | 4.83 ± 0.43 | 3.01 ± 0.27 | 2.98 ± 0.28 |
| BrainIAC | 0.744 ± 0.034 | 0.810 ± 0.031 | 2.61 ± 0.07 | 13.96 ± 0.59 | 3.59 ± 0.49 | 4.16 ± 0.25 |
| Dinov3D-neuro | 0.696 ± 0.041 | 0.757 ± 0.045 | 1.96 ± 0.11 | 8.29 ± 0.57 | 3.23 ± 0.27 | 3.69 ± 0.12 |
| FS aparc | 0.740 ± 0.034 | 0.770 ± 0.023 | 2.13 ± 0.17 | 9.09 ± 0.63 | 3.51 ± 0.31 | 3.99 ± 0.36 |
| FS Schaefer | 0.736 ± 0.028 | 0.740 ± 0.016 | 2.16 ± 0.19 | 8.29 ± 0.38 | 3.44 ± 0.31 | 3.99 ± 0.35 |
| SwinBrain | 0.744 ± 0.038 | 0.855 ± 0.023 | 2.39 ± 0.11 | 9.47 ± 0.54 | 3.41 ± 0.36 | 3.97 ± 0.20 |
| un-CNN † | 0.772 ± 0.043 | 0.857 ± 0.032 | 1.42 ± 0.08 | 5.11 ± 0.22 | 2.70 ± 0.33 | 3.13 ± 0.11 |
† un-CNN: Maintainers' own model.
| Model | Preprocessing | Dim | Links |
|---|---|---|---|
| 3D-Neuro-SimCLR | turboprep | 512 | paper |
| AnatCL (Global) | cat12 | 512 | paper · code |
| AnatCL (Local) | cat12 | 512 | paper · code |
| Brain2Vec | Own pipeline: NiChart DLMUSE (DLICV skull stripping) + FSL FLIRT to the Brain2Vec template | 768 | paper · model card · code |
| BrainIAC | turboprep | 768 | paper · code |
| Dinov3D-neuro | turboprep | 792 | model card |
| FS Schaefer | FreeSurfer v6.0.1 (RBC): Schaefer 400 cortical thickness and surface area | 800 | paper |
| FS aparc | FreeSurfer v6.0.1 (RBC): aparc 68 cortical thickness and surface area | 136 | paper |
| SwinBrain | turboprep | 384 | paper |
| un-CNN | turboprep | 9664 | paper · model card · code |