Doctors and computers are engaged in a race with no holds barred and that does not foresee "defeated": indeed, we will all gain in prevention. In fact, artificial intelligence overcomes dermatologists in distinguishing benign moles from melanomas : they are the conclusions of a study, published in the Annals of Oncology and reported by the Guardian, which opens positive scenarios on the future of early diagnosis.
The training. A group of researchers from Germany, the United States and France "taught" an automatic learning program - a convolutional neural network, ie software whose organization is inspired by that of the neurons of the animal visual cortex - to distinguish dangerous skin lesions from the harmless ones, starting from 100 thousand photographs.
Machine learning software cannot "see" like us, but learn, through exposure and repetition of stimuli, to distinguish recurring patterns (patterns) in images, and to categorize them with ever greater precision.The important things to know about tanning and related risks
There is no tender. After training, the success rates in the diagnosis of the AI program were compared with those of 58 dermatologists from 17 different countries and of various experience. Most doctors have been overtaken by software: human dermatologists have correctly identified 86.6% of the melanomas from images, while the software has hit the mark in 95% of cases .
The system has therefore "missed" less melanomas, making it therefore more sensitive than human sight, and has even exchanged less benign for dangerous injuries - a result that, in reality, would lead to less recourse to unnecessary interventions.
The context helps. The performance of dermatologists has improved - reaching 88.9% of the recognized melanomas - when additional information was provided with respect to a simple photograph, such as the patient's age or the position of the lesion (ie in conditions more similar to those of a normal preventive visit). However, the software has nevertheless overcome them, and starting from an image only.How are moles formed?
Let's join forces. The study does not mean that tomorrow skin doctors will be outclassed by virtual doctors: there is nothing better than an appointment with a careful and scrupulous dermatologist. In the future, however, in the trained eyes of doctors they may be supported by programs that make diagnosis even more accurate.