Colorist's color-matching trick boosts medical AI accuracy by 13%
A training-free color-matching method beats deep generative models on 4 medical datasets...
The authors introduce Colorist, a training-free, fully interpretable data augmentation strategy that repurposes classical statistical color matching—global mean-standard deviation matching in RGB space—for medical image classifiers. It safely generates structurally intact domain variations, outperforming deep generative models in structural fidelity and color alignment. Across out-of-distribution histopathology, peripheral blood, dermatology, and retinal datasets, it improves balanced accuracy by up to +9% over state-of-the-art domain generalization regularizers and by up to +13% over an unaugmented baseline. And by avoiding neural networks in the augmentation loop, Colorist preserves anatomical structure, minimizes carbon footprint, and integrates seamlessly into standard dataloaders—making statistical color matching a safe, interpretable, overlooked alternative to deep architectures for clinical robustness.
- Colorist improves balanced accuracy by +13% over unaugmented baselines and +9% over SOTA domain generalization regularizers across 4 medical imaging datasets
- Training-free classical statistical matching preserves anatomical structure and avoids hallucinations from deep generative style transfer
- Integrates into standard dataloaders, cutting compute and carbon footprint for clinical robustness
Why It Matters
A cheaper, safer way to make medical AI robust across hospitals, slashing compute costs without sacrificing accuracy.