1. Image capture:
Users are guided through capturing a high-quality, standardized image of their skin lesion using the app’s smart camera interface.
2. Initial risk assessment (within seconds):
The app provides one of three outcomes:
- • High risk: advise to see the doctor
- • Low risk: No immediate medical concern
- • In review: Image sent to dermatology panel for further assessment
3. Quality Assurance
All High risk and In review, and a small portion of Low Risk cases undergo secondary review by SkinVision’s panel of dermatologists to ensure accuracy and reliability.
Since 2012, SkinVision has continuously evolved its proprietary AI to improve diagnostic performance in real-world scenarios.
Model Architecture
- • ResNET-50 convolutional neural network – excels at analysing visual patterns.
- • The SkinVision app uses three slightly different variations of this network for more consistent results.
Model Training and Validation
- • Trained on a dataset of 200,000+ dermatologist-labelled images from diverse demographics and environments
- • Covers a broad variety of people and conditions, including different ages, genders, skin types, smartphone models and regions.
Model Testing
- • Independent bench test sets, performed with every model update, are used to test the model on images not used during training.
- • Performance validated using a dataset separate from the training data
- • Results from latest bench test:
91–96%
Sensitivity
Ability to correctly identify malignant lesions
89%
Specificity
Ability to correctly identify benign lesions
98.1%
Consistency
Stable outputs on similar images
Model Maintenance
- • Updated regularly with new data
- • Externally validated through clinical studies
- • Continuously monitored internally
Model Regulation
The SkinVision App is a class IIa medical device under the EU Medical Devices Regulation (MDR).It was the first AI-powered medical device to achieve CE registration under the EU Medical Device Directive (MDD) and currently has achieved the higher Class IIa certification under the MDR(EU MDR 2017/745).

