Researchers Develop New AI Tool That Makes Detection of Skin Cancer More Accurate

PHILADELPHIA (October 30, 2025) - Researchers at Fox Chase Cancer Center, Temple University's College of Engineering, and the Lewis Katz School of Medicine at Temple University have developed a new method that enhances the ability of artificial intelligence models to detect and diagnose skin cancer in individuals with darker skin.

"The biggest issue with current AI cancer detection models is that they are more effective at detecting melanoma in lighter skin tones and often have difficulty detecting it in darker skin tones. As a result, when melanoma is detected in patients with darker skin, those patients tend to be diagnosed at later stages" said Hayan Lee, PhD, corresponding author on the study, Assistant Professor in the Nuclear Dynamics and Cancer Research Program, and member of the Cancer Epigenetics Instituteat Fox Chase.

According to the researchers, existing AI models are not as effective at detecting melanoma in dark skin because of the kinds of data used to train them. This data often comes from just a few places and time periods, many times in one country, and doesn't represent all types of patients. When this occurs, detection methods can become biased, resulting in the AI tool diagnosing skin cancer more accurately in people with lighter skin tones than in people of color.

Recent studies on skin cancer diagnosis have looked at using advanced imaging and AI to improve detection for different skin types. However, most of these studies focused on how well the technology works and did not directly consider how skin color affects the results.

"There's this desire to have one big model, hoping that it can work for every skin type. I think this approach may be too general. It's important to understand and lessen the errors related to detection and skin types to create fair and accurate detection tools for everyone," said Lee, whose work highlights her background in computer science and engineering as well as her research interests in AI, computational epigenetics, computational oncology, and machine learning.

The lead author Vahid Khalkhali, a doctoral student in the Electrical and Computer Engineering Department at Temple, was co-advised by Lee and Saroj K. Biswas, PhD, Professor Emeritus at Temple. Researchers from the Katz School of Medicine and Ellis Monk, PhD, from Harvard University also contributed.

To develop a more accurate model, the research team developed a new method based on the Monk Skin Tone (MST) scale, a 10-shade scale designed to represent a more inclusive range of human skin tones. The researchers used a new method based on the MST scale called MST-AI to estimate skin color. The MST-AI method was then tested on a large public collection of skin cancer images.

"Our results show that MST-AI gives more accurate and reliable skin tone estimates than the other methods, based on trusted evaluation scores. It helps correct skin tone imbalances in large dermatology datasets, creating a better base for accurate and fair diagnosis," said Lee.

The team's novel method means doctors and patients can expect smarter tools that see beyond one-size-fits-all solutions. By making sure AI has data on a wider range of skin tones, this research aims to close the gap in skin cancer detection and provide earlier, more accurate diagnoses for everyone.

The study, "MST-AI: Skin Color Estimation in Skin Cancer Datasets," was published in the Journal of Imaging.