AI could exacerbate racial disparities in dermatology, researchers say

Artificial intelligence algorithms developed for dermatology must be created with "inclusivity in mind," according to an op-ed published in JAMA Dermatology.

In recent years, researchers have demonstrated how algorithms built using machine learning, a type of AI in which a computer learns over time, can detect malignant melanomas with similar accuracy to board-certified dermatologists. To train a machine learning algorithm, a research team must feed the program hundreds of data points, from which it can deduce relevant patterns.

These programs could potentially support dermatologists in early detection of skin cancers. However, the op-ed authors — Adewole S. Adamson, MD, a dermatologist at University of North Carolina at Chapel Hill, and Avery Smith, an engineer at computer software company Fearless in Baltimore — questioned whether existing datasets will exacerbate racial disparities.

"Although there is enthusiasm about the expectation that [machine learning] technology could improve early detection rates, as it stands it is possible that the only populations to benefit are those with fair skin," they wrote.

An algorithm develops its responses based on the data it's trained on, and this training is particularly important in a field like dermatology, where diseases may present with different features on darker skin types. Yet, most algorithms are trained on lighter skin types: For example, the International Skin Imaging Collaboration: Melanoma Project has primarily collected images from the U.S., Europe and Australia.

"Because [machine learning] is only as good as the data from which it learns, great care must be placed in building technology that reflects the diversity of skin types in our society," the authors wrote. "No matter how advanced the [machine learning] algorithm, it may underperform on images of lesions in skin of color."

More articles on artificial intelligence:
Nuance adds Mark Sherwood as CIO: 4 things to know
Harvard, UPenn researchers use AI to predict mortality for cancer patients
12 healthcare use cases for natural language processing

© Copyright ASC COMMUNICATIONS 2020. Interested in LINKING to or REPRINTING this content? View our policies by clicking here.

 

Featured Webinars

Featured Whitepapers