Methodology summary
Dermela should publish training data sources, model versioning, image-quality controls, evaluation sets, and exclusion criteria before making accuracy claims.
Science
AI methodology and limitations
This page keeps Dermela honest: publish validation before claims, explain limits clearly, and route concerning spots to medical care.
Accuracy reporting
Sensitivity, specificity, confidence intervals, population limits, and failure modes should be shown together. A single headline accuracy number is not enough.
Explicit limitations
Dermela is not medical advice. It does not diagnose melanoma, rule out cancer, or replace clinician assessment, dermoscopy, biopsy, or emergency care.
How Dermela frames AI skin insights
Dermela uses the phrase AI skin insights because the product is meant to support visual tracking and education. It should not be framed as a diagnostic device, a clinician substitute, or a substitute for clinical judgement.
Any future performance claims should describe the exact model version, image-quality thresholds, data collection process, excluded images, evaluation population, and clinically relevant failure modes. Accuracy without those details can be misleading, especially across different skin tones, body areas, lighting conditions, and lesion types.
What belongs in validation
A useful validation summary should report sensitivity, specificity, positive and negative predictive values where appropriate, confidence intervals, and the reference standard used for labels. It should also explain what Dermela does when an image is blurry, poorly lit, cropped, or otherwise unsuitable for review.