Get Started
Home
Topics
Search
Library
Research questionWhich shift most undermines dermatology AI generalization: skin tone or disease distribution?Dermatology models trained on light-skinned, cancer-focused image collections may encounter both different skin tones and unfamiliar clinical conditions after deployment. Because these shifts can be confounded, a performance drop does not reveal which source of mismatch is responsible.
AI
Computer Vision
Evaluation & Benchmarks
Health
Machine Learning
Research Paper
Latest papersRecent research connected to this question, newest first.Disease Burden over Skin Tone: Decomposing the Dermatology-AI Generalization GapThe study evaluates a cancer-trained ResNet-50, two dermatology foundation models, and a general vision model as frozen feature extractors on a tone-stratified, disease-matched dataset and a tone-diverse, disease-shifted dataset. In the evaluated settings, disease-distribution shift had a larger effect than skin tone; the evidence is limited to the tested models, datasets, and adaptation settings.research paper · Sep 2, 2026
Related questions
How can AI-generated text detectors remain reliable under distribution shift?How can retinal fundus models remain interpretable and accurate when pathology alters vessel appearance?When does clinical text materially influence pixel-level predictions in medical image segmentation?How can pediatric chest X-ray classifiers preserve usable decisions across countries when calibration and source thresholds shift?