Researchers have introduced a new AI method called MST-AI that aims to make skin cancer detection more accurate and fair for people with darker skin tones. Many current AI tools used to detect melanoma work well on lighter skin but often fail on darker skin because the training images they rely on are not diverse enough. This imbalance can lead to later diagnoses and worse outcomes for patients of color.
When the images used to train these models are mostly of lighter skin, the AI is more likely to make mistakes when evaluating darker skin tones. To address this, the research team created MST-AI, a method that estimates skin color more accurately in large dermatology datasets.
MST-AI is based on the Monk Skin Tone scale, a 10-shade system designed to reflect a broader and more inclusive range of human skin colors than older classification scales. By applying this method to a large public collection of skin cancer images, the team found that MST-AI provided more precise and consistent skin tone estimates than other existing techniques.
This improvement helps correct skin tone imbalances in dermatology datasets. With more accurate labeling and a fairer distribution of skin tones, future AI models can learn from data that better represents all patients. This, in turn, supports more reliable melanoma detection across different skin types and helps close the diagnostic gap that has long affected people with darker skin.