How the AI works
Transparency matters to us - especially in a health context. Here we explain honestly how your skin analysis is produced, what the AI can do, and where its limits lie.
The flow in brief
1. Photo
You photograph the skin spot with your phone or upload an image. The image is health data and is processed only with your explicit consent.
2. Analysis
Several AI models independently assess the image for features that may indicate benign or malignant change.
3. Assessment
You receive an understandable risk assessment with an ABCDE breakdown and a clear recommendation on whether, and how urgently, to see a doctor.
Why several models instead of one
A single model can be wrong. That is why DermCheck combines six specialised, openly available image-classification models (trained on dermatology datasets) with a large vision-language model that assesses the spot structurally along the ABCDE rule and a 7-point checklist. The individual results are merged with weights. When the models disagree or are uncertain, we present the result more cautiously and lean towards recommending a medical check - when in doubt, for safety.
Uncertainty is measured too
The AI provides not just a result but also a measure of how confident it is. If the system detects that an image is unusual or is not a typical skin lesion at all (out-of-distribution), or if the models diverge strongly, the result is shown deliberately conservatively and a medical check is advised.
What the AI can NOT do
- It does not make a diagnosis and is not a medical device.
- It does not replace a medical examination, dermoscopy, or biopsy.
- A benign result is no guarantee - for persistent or changing spots, always have them checked.
- Image quality, lighting, and skin type affect the result.