AI-Assisted Skin Cancer Screening and Diagnostics
AI and machine learning methods for earlier and more accurate diagnosis of skin cancers and disorders from smartphone images
Melanoma and other skin cancers are among the most common cancers globally, yet access to dermatology care remains highly unequal — with long wait times and geographic barriers affecting timely diagnosis. Smartphone-based AI diagnostics offer a potential path to democratizing access to expert-level skin assessment.
The challenge. Real-world smartphone images of skin lesions are noisy, variable in lighting, angle, and resolution, and often captured by untrained users — making robust, clinically reliable AI diagnosis significantly harder than performance on curated benchmark datasets.
Our approach. In collaboration with Dr. Maria Wei (UCSF Dermatology and San Francisco VA), we are developing AI algorithms for robust home-based skin lesion diagnostics that:
- Handle the full distribution of real-world image quality and acquisition conditions
- Provide calibrated uncertainty estimates alongside diagnostic predictions
- Are evaluated prospectively against dermatologist assessments
Context. This work is part of a broader initiative to develop reliable, equitable AI diagnostic tools that can be deployed in primary care and direct-to-patient settings, with particular attention to underserved populations served by the SF VA.