AI-Assisted Melanoma Rule-Out

Clinical Decision Support for Limited-Resource Settings

When a primary care physician sees a suspicious skin lesion, the choice matters. Refer to a specialist — and a patient in a rural or underserved community faces travel costs, time away from work, and a wait that may stretch months. Don't refer — and a melanoma may be missed.

This AI tool is designed to help make that decision with confidence. When it indicates low risk, it is correct more than 96 times out of 100.

The Evidence

Trained on 16,128 histopathology-confirmed dermoscopic images from multiple institutions worldwide, validated on 562 independent cases:

AUC-ROC: 0.982 (95% CI, 0.967–0.997)

Specificity: 96.1% (95% CI, 93.8%–97.6%) — correctly clearing benign lesions

Sensitivity: 90.3% (95% CI, 84.6%–94.0%) — detecting melanoma

NPV: 96.3%–99.9% across realistic prevalence ranges

NNI: 1.11 — only 1.11 referrals per melanoma detected

Unlike currently authorized AI skin cancer detection devices reporting specificity of 20.7%–32.5%, this model achieves 96.1% specificity — substantially reducing unnecessary referrals without compromising melanoma detection.

Access the Demo →

Research and Educational Use Only — Not FDA Authorized. This tool is not intended for clinical diagnosis or treatment decisions. Results must be confirmed by qualified dermatologists. For research, education, and demonstration purposes only.

Model trained on ISIC Archive dermoscopic images (CC-BY licensed). Performance in darker skin tones requires dedicated validation before deployment in racially diverse populations.