Crowdsourcing Dermatology Images With Google Search Ads: Creating A Real-World Skin Condition Dataset

    February 2024 in “ arXiv (Cornell University) ”
    Abbi Ward, Jimmy Li, Julie Wang, Sriram Lakshminarasimhan, Ashley Carrick, Bilson Campana, Jay Hartford, Pradeep Kumar S, Tiya Tiyasirichokchai, Sunny Virmani, Renee Wong, Yossi Matias, Greg S. Corrado, Dale R. Webster, Dawn H. Siegel, Steven Lin, Justin Ko, Alan Karthikesalingam, Christopher Semturs, Pooja Rao
    TLDR Google Search ads effectively gathered a diverse dermatology image dataset for research and AI development.
    The study demonstrates the effectiveness of using Google Search ads to crowdsource a diverse and representative dataset of dermatology images, addressing gaps in real-world health data. Over 8 months, 10,408 images were collected from 5,033 contributors in the US, with a higher representation of females and younger individuals. The dataset includes dermatologist-labeled conditions and skin tone estimates, with most images depicting allergic, infectious, or inflammatory conditions. The SCIN dataset, available online, enhances research and AI tool development by providing a broad spectrum of skin condition images.
    Discuss this study in the Community →