The AI model makes these predictions based on a facial photo. The developed AI algorithm, according to research by Erasmus MC, can ‘see’ characteristics of skin cancer that dermatologists – who pay attention to classic characteristics such as pigment spots, wrinkles or the fact that someone has lived in a sunny country – cannot detect. .
The AI model works better than the current way in which doctors make a risk assessment. The research was led by dermatologist Dr. Marlies Wakkee, together with AI expert Dr. Gennady Roshchupkin. “The algorithm picks up more characteristics than just the classic risk factors for skin cancer,” says Dr. Wakkee.
Training AI model
More than 2,800 facial photos were used to train the AI model and algorithm. Those photos were all taken under the same conditions. The researchers knew which people whose facial photos were taken developed skin cancer. The algorithm is able to extract 200 characteristics from the photos. Based on these characteristics, the AI model can then predict the risk of developing skin cancer, both on the face and other places on the body.
This prediction, the research shows, is more accurate than a prediction that can only be made on the basis of a standard skin examination with questionnaires and/or genetic research. The results of the study have been published in the scientific journal The Lancet’s eClinicalMedicine.
Explainable AI methods
So-called explainable AI methods were used for the research. This makes it possible to find out exactly what the algorithm’s predictions are based on. This is not always the case with AI applications. In those cases, it is not clear how an AI application arrived at a certain outcome.
“We make visible the 200 characteristics on which the model bases its prediction. This also includes characteristics that are less clear to people,” says Dr. Wakkee. “Explainable is not the same as interpretable. We can explain what the algorithm does, but what its meaning is – in a human way – is not so clear,” Roshchupkin adds.
Future roles
Wakkee and Roshchupkin see two possible roles for the AI model that predicts the risk of skin cancer in the future: prevention and awareness. “If you know from a photo that you have a high risk of skin cancer, you use the sun more consciously. We also hope that warned people will keep a closer eye on their skin,” the researchers said.
It will take some time before this AI model can actually be applied in practice. A number of follow-up studies still need to be carried out for this purpose. For example, the model still needs to be tested on larger datasets, where facial photos of people of more different ages and ethnic backgrounds must also be used. The researchers have made the model publicly available to the research community with the aim of accelerating the entire AI community.
A milestone was achieved in the development of the application of AI for the diagnosis of skin cancer in the US in January this year. The FDA then approved the use of a wearable AI derma sensor for use in clinical practice. This sensor uses artificial intelligence (AI) to analyze the reflection of light on skin lesions and assesses whether it is a benign or malignant spot based on light patterns.
