AI for skin cancer screening

Since October, in Evreux, a town located around a hundred kilometers west of Paris, a huge machine 3 m high and 5 m wide – Vectra 360 – has been capable of photographing almost the entire surface of the skin of a person. a patient in one shot, thanks to its 92 high definition lenses. The scanner produces a map of visible lesions and all moles. “The doctor can then, remotely, choose to enlarge one that seems suspicious to analyze it,” explains Isabelle L’Hôpital, who directs the company France Dermatologie Territoires, at the origin of the project.

The objective of this dermatological imaging sector is to improve the detection of melanoma in a department which faces, like many others, “long delays in obtaining an appointment with a dermatologist” on background of shortage of these specialists. Each year in France, around 18,000 cases of melanoma are discovered, causing 2,000 deaths.

Melanoma is a skin tumor that looks like a mole but often has the following characteristics: asymmetry, irregular edges, multiple colors, enlargement, or change in appearance. The number of new cases per year has seen a steady increase over the past two to three decades. But thanks to improved screening and the introduction of new treatments, the death rate has tended to stabilize in recent years. Several manufacturers are trying to revolutionize the early treatment of this serious cancer.

“Since the end of the 1990s, many practices have been equipped with so-called digital dermoscopy machines, which make it possible to take photos of a patient’s skin and then compare them during successive appointments to see if the grains of beauty have evolved,” says Luc Thomas, specialist in skin cancers, practitioner at Lyon University Hospital.

But technologies are evolving: the quality of images has significantly improved. Previously capable of photographing skin cm2 by cm2, machines can now capture almost the entire body surface. The French startup SquareMind hopes to market its “innovative” solution this year: a robot arm that navigates around the patient. “It will offer in just a few minutes a photograph of the entire body, zoomable on the lesions up to very high definition,” explains Ali Khachlouf, founder of SquareMind.

At a time when the dermatological offer is reduced, automating certain time-consuming steps of melanoma screening will make it possible to follow more precisely the patients who need it most, explains Jilliana Monnier, oncodermatologist, head of the automated melanoma screening center in Marseille.

Secondly, practitioners hope to be able to count on the help of artificial intelligence, produced by these machines. By automatically documenting the skin surface of a large number of people, they will make it possible to create a complete history of these patients’ lesions and moles. The objective will then be to develop AI algorithms capable of easily and quickly identifying new, progressive or suspicious lesions on the skin as a whole.

“Today, the machine cannot yet give a diagnosis,” explains Jilliana Monnier. That is to say, confirm whether this or that lesion is a melanoma. But within two years, Ali Khachlouf expects an AI that detects melanoma “with a high level of confidence”. “Artificial intelligence could be used to do a sort of sorting, even if the final decision will always fall to the doctor,” says Luc Thomas.