A team of researchers from Fox Chase Cancer Center and Temple University has developed a new way to improve how artificial intelligence (AI) detects skin cancer, especially in people with darker skin tones.
This work could lead to earlier and more accurate diagnoses for everyone, no matter their skin color.
The research, published in the Journal of Imaging, shows how a new AI tool called MST-AI can help fix one of the major problems with current skin cancer detection models. Most existing AI models are better at spotting melanoma—the most dangerous type of skin cancer—in people with light skin.
But they often miss or delay detection in people with darker skin, which can lead to more serious cases by the time they are diagnosed.
Dr. Hayan Lee, one of the lead authors, explained that this problem exists because most of the images used to train these AI tools come from only a few countries and mostly show light-skinned patients. This creates a bias in the AI models, making them less accurate for people with darker skin.
While many recent studies have looked at how to improve AI for diagnosing skin cancer, most of them don’t focus directly on skin color. Instead, they test how well the AI works overall without paying attention to the differences in skin tones. This general approach doesn’t fix the problem of fairness and can leave people of color at a disadvantage.
Dr. Lee, who works in computer science and cancer research, said it’s not enough to try to make one model fit all. Instead, we need to understand and reduce the errors caused by differences in skin color. That way, AI tools can work well for everyone, not just some people.
The lead researcher, Vahid Khalkhali, a Ph.D. student in engineering at Temple University, worked with a team of scientists from Temple, Fox Chase, and Harvard University. Together, they built the MST-AI tool using something called the Monk Skin Tone (MST) scale. This scale has 10 shades and is designed to better reflect the full range of human skin tones.
The MST-AI method was tested on a large public database of skin cancer images. It was found to be more accurate than older methods at estimating skin tone. This helped the researchers balance the skin tone data in their image collections, which is an important first step toward making AI detection tools more fair and reliable.
The researchers say that by using MST-AI, future AI models for skin cancer will be smarter and more inclusive.
Doctors will be able to diagnose cancer earlier, and patients from all backgrounds will receive better care. This is especially important for people who have been left out of previous research and whose cancers may have gone undetected until later stages.
This study represents a major step forward in using technology to fight health inequalities. By making sure AI tools include data from people of all skin colors, the team hopes to close the gap in skin cancer detection and help save lives.
If you care about cancer, please read studies about common drugs for inflammation may help kill cancer, and statin drugs can starve cancer cells to death.
For more information about cancer, please see recent studies about these two things are key to surviving cancer and results showing common Indian fruit may slow down cancer growth.
The study is published in the Journal of Imaging.
Copyright © 2025 Knowridge Science Report. All rights reserved.
