A general symptom triage engine and a dermatology-specific model serve different purposes, even when both analyse photographs. A general triage engine is designed to assess a broad range of symptoms and help guide the appropriate level of care. A dermatology-specific model, such as Autoderm, is focused specifically on skin presentations and is developed and validated around the characteristics, conditions and variations that are relevant to dermatology.
This difference matters when evaluating a tool in practice. Evaluators should look beyond the interface to understand how the model was trained and validated, what data support it, and what it is designed and permitted to do.
Therefore, evaluating a dermatology-specific model requires looking beyond whether it can produce an image-based result. This article highlights the key factors technical evaluators should consider when assessing a dermatology AI vendor.
A dermatology-specific model takes a different approach. Its training data, condition categories and validation are focused specifically on skin presentations. By concentrating on one clinical area, the model can develop greater depth in recognising and differentiating skin conditions. This difference is important when evaluating a tool intended specifically for dermatology, rather than general symptom triage.
Skin-tone representation in the training and validation data should also be considered when assessing a dermatology model. It is important to look beyond a general statement that a model has been “validated” and understand which skin tones are actually represented in its data and testing. This makes skin-tone coverage a relevant purchasing and evaluation criterion, rather than simply a marketing claim.
What is the device’s current regulatory classification, what is its stated intended use, and does this match how the product is currently marketed?
Is the product currently CE marked, and if so, which notified body was involved in the conformity assessment?
What skin tones were included in the model’s training and validation, and can the validation results be provided by Fitzpatrick skin type?
How many skin conditions can the model suggest, and how is this number defined?
What peer-reviewed publications and post-market clinical follow-up studies are available to support the model’s performance, beyond the vendor’s own internal testing?
What does the product’s post-market surveillance record show to date, including any reported performance, safety or usability issues, and can this evidence be shared?
What image formats, resolution, file-size limits and other image-quality requirements does the system have, and how does it handle images that do not meet those requirements?
What is the typical response time and what integration options or technical formats are available?
Where are submitted images and related data processed and stored, and what data protection requirements does the system support?
Each of these areas should be supported by a specific document, test result, publication, certification or measurable figure wherever possible. A general assurance statement is not enough on its own.
A vendor confident in its own evidence base should be able to answer all nine without hesitation.
Autoderm is CE-marked as a Class I medical device (MDD), with its MDR Class IIa application in progress. It is also registered as a medical device with the MHRA in the UK , and holds FDA Breakthrough Device Designation, with clearance pending in the US. Full intended-use documentation is available in Autoderm’s instructions for use .
Full regulatory status is available on Autoderm’s regulatory page.
Teams evaluating Autoderm against the criteria above can also start for free . Creating an account automatically assigns Free tier membership, which includes five free API calls to try the system for themselves before making a decision. Full pricing details are available on the pricing page .
