Improving data participation for the development of artificial intelligence in dermatology

Abstract

Artificial intelligence (AI) has the potential to significantly impact many aspects of dermatology. The visual nature of dermatology lends itself to innovations in this space. The robustness of AI algorithms depends on the quality, quantity, and variety of data on which it is trained and tested. Image collections can suffer from inconsistencies in image quality, underrepresentation of various anatomic sites and skin tones, and lack of benign counterparts leading to underperformance of algorithms in contexts other than one in which it is developed. Access to care, trust, rights, control, and transparency all play roles in the willingness of patients and health care providers and systems to collect, provide, and share data. Opportunities to improve data participation for the development of AI include the establishment of data hubs and public algorithms, federated learning strategies, development of renumeration ecosystems for patients and systems, and development of criteria and mechanisms for transparency.