Google’s AI can recognize 26 skin conditions as well as dermatologists

Google has developed a tool that can detect a skin condition based on images. The smart software could diagnose 26 different skin conditions just as well as dermatologists.

Developers within Google have created a deep learning tool about which they have written a publication, which has been made accessible via the arXiv science archive. In a blog post, Google points out that there are 1.9 billion people worldwide with a skin condition, but not all of them are treated by a dermatologist. Because in many cases the treatment is done by a non-specialist GP, in many cases the diagnosis is incorrect, according to Google.

The tool that Google has developed could make it easier to make the correct diagnosis if a dermatologist is not available. The software makes it possible to make a diagnosis based on images of the patient’s skin and available metadata. 17,777 cases were used to train the system, linked to the diagnoses of 40 different dermatologists.

Just like doctors often do, the AI ​​makes a differential diagnosis with a sort of ranking of possible conditions that belong to a certain picture. Comparisons show that the diagnoses provided by the software are about as accurate as those provided by dermatologists. That was the case with 26 different skin conditions. Incidentally, skin cancer turned out to be more difficult to detect, because there are fewer cases of it and the system could therefore be trained less well.

Although Google believes the artificial intelligence can be used to help diagnose skin conditions, improvements need to be made in the future to be able to diagnose less common conditions as well. By the way, algorithms have already been developed that can recognize skin cancer just as well as dermatologists.