Related Experiment Video
Updated: Feb 19, 2026
08:05
Detection and Isolation of Cancer in Prostate Biopsies Using Stimulated Raman Histology and Artificial Intelligence
Published on: June 10, 2025
1.3K
Artificial intelligence-based smartphone application for skin cancer detection: a prospective diagnostic accuracy
Julie Kips1,2,3, Jorien Papeleu1,2,3, Amber Shen1,2,3
1Cancer Research Institute Ghent (CRIG - SkinCRIG), Ghent, Belgium.
The British Journal of Dermatology
|February 17, 2026
Summary
This study found that an AI skin cancer detection app had moderate accuracy, with performance varying by image quality and phone model. Independent validation is crucial for AI healthcare tools.
Area of Science:
- Dermatology
- Artificial Intelligence
- Medical Imaging
Background:
- Smartphone applications utilizing artificial intelligence (AI) show promise for early skin cancer detection.
- Prospective studies evaluating the real-world performance of these AI tools are limited.
Purpose of the Study:
- To independently assess the diagnostic accuracy of a popular AI skin cancer detection app.
- To investigate the impact of photographic conditions on the app's performance in a diverse patient group.
Main Methods:
- A prospective study enrolled 1,458 participants with 1,904 lesions of concern.
- Lesions were photographed using the AI app, and a convolutional neural network (CNN) provided risk assessment.
- Diagnostic accuracy was compared to clinical/histopathological diagnoses, with performance analyzed under varied photographic conditions and smartphone models.
Main Results:
- The AI app achieved 82.5% sensitivity and 76.8% specificity for skin cancer detection on successfully captured images.
- Image capture failed in 16.6% of lesions. Combined AI and teledermatology review improved specificity to 86.8%.
- Performance varied significantly by smartphone model and user image capture success (28.9%).
Conclusions:
- Independent clinical validation of AI healthcare tools is essential for real-world application.
- The study underscores the need for robust testing of AI diagnostic tools across diverse conditions and devices.
