Melanoma is a type of cancer that affects pigment-producing skin cells; it kills roughly 7,000 men and women in the United States each year. Early detection and treatment of melanoma give patients the best chance of surviving the disease. Researchers from the Wyss Institute for Biologically Inspired Engineering at Harvard University and the Massachusetts Institute of Technology (MIT) have developed a novel computer-assisted diagnosis (CAD) system that screens skin lesions for melanoma using convolutional deep neural network technology.
Human dermatologists use a technique called the “ugly duckling” method to assess skin lesions. This involves observing multiple skin moles on a single patient. When dermatologists find an “ugly duckling” — a lesion that looks different from the others — they examine it for cancer markers. When dermatologists identify a suspicious pigmented lesion (SPL), they typically recommend a biopsy to confirm the diagnosis.
Unlike previous CAD systems that evaluate only the target skin lesion, the new CAD system essentially replicates the “ugly duckling” process. The research team trained a deep learning neural network to analyze photos of a patient’s skin based on more than 33,000 images. This image database includes dermatologist-confirmed SPLs as well as lesions that appear normal. The team also trained the algorithm to create a 3D map of all lesions within the patient photo and evaluate them for signs of abnormality along with the target lesion.
The researchers asked three experienced dermatologists to assess photos from 68 patients, then compared their evaluations to the CAD assessments of the same photos. The neural network agreed with the consensus of all three physicians 88% of the time, and with at least one dermatologist 86% of the time.
The research team has made the algorithm available on the open-source platform GitHub. That means that anyone with a smartphone could eventually use the tool to screen photos of concerning skin lesions, eliminating the costs and time associated with a preliminary dermatology exam. That could make early detection and treatment of melanoma more accessible for a significant number of adults, potentially changing the survival rate for a deadly form of cancer.