AI Tool Improves Accuracy of Skin Cancer Detection

Many existing artificial intelligence (AI) tools detect skin cancer more reliably in lighter skin tones, often missing early signs in patients with darker skin. This contributes to later-stage diagnoses and poorer outcomes. To address this diagnostic gap, researchers have developed a new AI-based method that improves skin tone recognition and enhances melanoma detection across diverse patient groups.

The collaborative research led by Fox Chase Cancer Center (Philadelphia, PA, USA) introduces a more equitable approach to training AI tools for dermatology. The researchers focused on understanding why existing AI systems underperform in darker skin tones. They found that most models are trained on narrow datasets—often sourced from limited geographic regions, hospitals, or time periods—and fail to represent the full spectrum of human skin color.

This lack of diversity leads to biased diagnostic results that favor lighter-skinned individuals. Recent imaging and AI studies have sought to enhance detection across skin types, but few have directly examined how skin color affects diagnostic performance. To overcome this limitation, the team developed MST-AI, a method based on the Monk Skin Tone (MST) scale, which uses 10 distinct shades to capture a wider range of skin tones. MST-AI estimates skin color more precisely and was tested on a large public dataset of skin cancer images.

The approach, highlighted in the Journal of Imaging, provides stronger accuracy and more reliable skin tone assignments compared with existing techniques, creating a more representative foundation for AI training. The MST-AI method helps correct skin tone imbalances in dermatology datasets, allowing AI models to learn from a more inclusive sample. This can improve early detection, reduce missed or delayed melanoma diagnoses in people of color, and support the development of fairer clinical AI tools.

“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 Hayan Lee, PhD, corresponding author on the study.

Related Links:
Fox Chase Cancer Center