Evidence-based by design

Our datasets

SkinGenius models are trained on more than 2.4 million dermoscopic and clinical images, curated from partner institutions and public research datasets. Every image is labeled against clinical ground truth — histopathology where available, or consensus review by board-certified dermatologists.

Validation methodology

We evaluate on held-out datasets that the model never sees during training, and benchmark against panels of dermatologists on the same cases. We report per-condition sensitivity and specificity — not just headline accuracy — so partners understand exactly where the model is strong and where human review matters most.

Fairness across skin tones

Dermatology AI has historically underperformed on darker skin. We deliberately balance training data and audit performance across all six Fitzpatrick skin types, treating any meaningful gap as a release blocker rather than a footnote.

Explainability

Each prediction is accompanied by saliency overlays that highlight the regions driving the result, a ranked differential, and calibrated confidence scores. Clinicians see the reasoning, not just a label.

Responsible AI

• Human-in-the-loop: outputs are decision-support, never autonomous diagnosis.
• Continuous monitoring for data drift and real-world performance.
• Clear communication of uncertainty and known limitations.
• Governance under an ISO 13485 quality management system.

Publications & collaboration

We work with academic and clinical partners on peer-reviewed validation studies. Detailed performance reports and study protocols are available to qualified institutions under a research agreement.

Interested in collaborating? Reach our research team at contact@skingenius.app.