Extended Data Fig. 2: Few-shot benchmarking of pretrained encoders.
From: A generalizable foundation model for analysis of human brain MRI
Few-shot performance benchmarking of three self-supervised pretrained MRI encoders—SimCLR-ViT-B 3D, SimCLR-ResNet50, and MAE-SwinViT—across seven downstream tasks evaluated with 1-shot (K1) and 5-shot (K5) training. Data are presented as mean values ± 95% confidence intervals estimated from 1,000 bootstrap samples. a Overall survival prediction (AUC), b MRI sequence classification (balanced accuracy), c brain age prediction (MAE in years), d IDH mutation prediction (AUC), e dementia prediction (balanced accuracy), f time-to-stroke prediction (MAE in days), and g tumor segmentation (mean dice).
