How Dermo's AI checks your skin

AI analysis in Dermo is an optional extra. The regular ABCDE check and tracking over time are free to use, while an AI analysis of a single photo costs $1 per analysis. This page explains how the AI actually works, and where it falls short.

Three models instead of one

When you order an AI analysis, Dermo doesn't ask a single model for the answer. Three independent AI models analyse the photo at the same time, each on its own, without knowing what the others think. Each one produces a list of possible conditions and a read on whether the spot looks benign or suspicious.

Once all three have given their initial answers, Dermo shows them each other's conclusions and asks them to look at the case again. This second round works a bit like three doctors comparing notes. If one of the models noticed something the others missed, the other two get a chance to revise their thinking. The final answer is a consensus built from all three revised assessments, which makes the result more reliable than if a single model had done the job alone.

What the AI is looking at

Dermo's AI is trained to think the way a dermatologist does during a visual examination. That means paying attention to the morphology of the lesion (whether it is flat, raised, or nodular), the colour and whether it varies across the surface, the texture (smooth, scaly, crusted, or warty), the distribution pattern if there are multiple lesions, and the location on the body, since some conditions tend to appear in particular places.

For pigmented moles, the AI is trained to think through the classic ABCDE criteria (asymmetry, border, colour, diameter, and evolution) as one of several lenses it uses when forming its assessment. For non-pigmented spots like warts, acne, eczema, or fungal infections, it leans more on morphology and pattern recognition. In both cases the analysis goes through the same three-model process to reach a shared conclusion.

Why you always see how sure the AI is

Every answer from the AI comes with a clear confidence score: how sure it is of its assessment. If the AI is confident, you'll see a high score and a clear recommendation. If it's genuinely uncertain, the score drops, and you'll see a note explaining what is unclear.

We show you this because hiding uncertainty would be dishonest. A medical tool that pretends to always know the answer is more dangerous than one that admits when it doesn't. A low-confidence result should be weighted differently from a high-confidence one.

What the AI cannot do

Dermo cannot replace a dermatologist, and we are not trying to. A smartphone photo gives the AI far less information than a clinician sitting in the same room as you has access to. A dermatologist can feel the texture, see the lesion under different lighting, ask follow-up questions on the spot, and use a dermatoscope to examine deeper structures that aren't visible in an ordinary photograph.

The AI also has blind spots. It performs best on typical lesions that look like the ones it has seen plenty of in its training data. Rare conditions, unusual presentations, and lesions in places that are hard to photograph (inside the mouth, under a nail, on the scalp) are harder for it to assess. Photo quality matters too: a blurry, dim, or poorly framed image will get a less reliable answer.