Employing Artificial Intelligence to Detect Skin Cancer Earlier

Although science fiction paints robots as taking over the world and defeating humanity, scientists are harnessing artificial intelligence (AI) to save lives.

Through a collaboration between the University of Waterloo and the Sunnybrook Research Institute, scientists developed a novel AI system trained using tens of thousands of skin images and their corresponding hemoglobin and eumelanin levels to detect melanoma, according to a press release.

The technology uses machine-learning software to analyze skin lesion images and provide physicians with data on telltale melanoma biomarkers, which is treatable if caught early.

“This could be a very powerful tool for skin cancer clinical decision support,” said investigator Alexander Wong. “The more interpretable information there is, the better the decisions are.”

Dermatologists largely rely on subjective visual examinations of skin lesions to determine if patients should undergo biopsies. With the novel technology, however, it could reduce the number of unnecessary biopsies by providing physicians with objective information on lesion characteristics that help rule out melanoma before taking a more invasive approach.

The system deciphers levels of biomarker substances in lesions and adds quantitative information to better assess the lesion.

Changes in the concentration and distribution of the chemical eumelanin is a particularly strong indicator of melanoma. When biomarkers of melanoma are detected early it is highly treatable, but after the disease has advanced, it becomes deadly.

“There can be a huge lag time before doctors even figure out what is going on with the patient,” Wong said. “Our goal is to shorten that process.”

Approximately 87,110 new cases of invasive melanoma will be diagnosed in the United States in 2017, with an estimated 9730 patients who will die from the disease. Although melanoma accounts for less than 1% of skin cancer cases, it accounts for most skin cancer deaths.

The study findings were recently presented at the International Conference on Image Analysis and Recognition in Montreal, Canada.