Medical Image Classification on OASIS (AD/CN) 1 (test)
78AccuracyMulti-Scale Reward Learning (Ours)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Multi-Scale Reward Learning (Ours)Backbone=3D ResNet-50, Data Source=RL fine-tuned diffusion model (2000 volumes per class), Training=Pre-trained on RL synthetic then fine-tuned on real data, Evaluation Protocol=Mean ± standard deviation over three independent runs2026.03 | 78 | 77 | 86 | |
| Standard SyntheticBackbone=3D ResNet-50, Data Source=Standard diffusion models, Training=Pre-trained on synthetic data then fine-tuned on real data, Evaluation Protocol=Mean ± standard deviation over three independent runs2026.03 | 77 | 75 | 80 | |
| Real Data OnlyBackbone=3D ResNet-50, Training=Fine-tuned on real training dataset, Evaluation Protocol=Mean ± standard deviation over three independent runs2026.03 | 76 | 75 | 81 |