(Photo by Flickr user janholmquist, used under a Creative Commons license)
Technical.ly’s Editorial Calendar explores a different topic each month. The August 2018 topic is Technologists of Color. These stories highlight the contributions of technologists and entrepreneurs of color across Technical.ly’s five markets.
As dermatologists look to use tech to better diagnose skin diseases, they’re beginning to turn to machine learning to help testing moles for cancer. But a new paper warns of the potential for racial disparities as a result of the data that’s at the heart of the algorithms.
Published in Jama Dermatology, a peer-reviewed journal published by the American Medical Association, the paper by Adewole S. Adamson and Baltimore software developer Avery Smith, warns of racial bias in machine learning that’s being used to distinguish between moles that are malignant and benign in the process of testing for melanoma.
The paper states that machine learning is promising for the field, with the potential to improve patient care. To do so, however, more diverse data is required in training algorithms so as not to make healthcare disparities worse. The co-authors point to one archive where most patient data is collected from fair-skinned patients in the United States, Australia, and Europe.
“…No matter how advanced the [machine learning] algorithm, it may underperform on images of lesions in skin of color,” the paper states. The differences include how a mole is analyzed in terms of the contrast of its color with a skin tone, as well location on the body, said Smith, of Baltimore dev shop Fearless.
“If your training is based on pathologizing a mole based on contrast and based on how it tends to present, then you’re going to always judge other moles that come to you on different kinds of skin in the same way,” Smith said.
Black patients experience higher mortality rates from melanoma than fair-skinned patients, the study states, and patients with skin of color often present at more advanced stages. Though the disease is more prevalent among caucasian patients, “that does not mean that patients of darker skin types should be excluded from potential benefits of early detection through [machine learning],” the paper states.
The paper calls for creating software that includes all skin types. Smith said he wants to collect data on skin of color in areas with predominantly black populations, and create a repository that could be used for such software.
Smith’s wife, LaToya Smith, passed away in 2011 as a result of skin cancer. A podiatrist who served on medical missions in Eritrea and a Native American reservation in Arizona,LaToya Smith entered private practice in 2010. Avery Smith maintains a website honoring her memory. By working on new advances, he is hoping to help people suffering from similar conditions in the future.
“My life changed when I became widowed and yet the impact of my relationship lives on,” Smith said.
This JHU team is developing tools to assess newborn health
Medical education startup Osmosis raises $4M Series A
CareFirst, LifeBridge Health plan innovation center, fund for digital health startups
How law firm Nemphos Braue is guiding startups along the new business learning curve
Baltimore’s Sonavi Labs takes third place at Startup World Cup
Osmosis reaches 1 million YouTube subscribers with approachable medical content
UMB startup CoapTech receives FDA clearance for feeding tube placement device
Building a data acquisition system? Don’t make this mistake
Sign-up for daily news updates from Technical.ly Baltimore