Google develops all kinds of AI tools for social and scientific purposes. In recent years, the company has tested the difference tools: for predicting floods, protecting whales and, for example, detecting skin conditions. Developers of Google AI have now created a deep-learning tool, which can be viewed via the scientific archive of arXiv. In the publication, we read that 1.9 billion people in the world have skin conditions, but not all of them are treated by a dermatologist.
Skin disorders are among the most common forms of illness, after colds, tiredness and headaches. Google estimates that around 25 percent of all patients worldwide are under treatment for skin conditions. Worldwide there is a shortage of dermatologists, which means that many GPs make diagnoses themselves and they are less accurate than when they are done by a specialist. Researchers at Google developed an AI system with which the most common dermatological disorders in primary care can be discovered. In the publication we read that it has proven to be accurate for 26 different skin conditions.
— VentureBeat (@VentureBeat) September 13, 2019
Potential of the DLS
“We have developed a deep learning system (DLS) to address the most common skin conditions in primary care,” wrote Google software engineer Yuan Liu and Google Health technical program manager Dr. Peggy Bui. “This study emphasizes the potential of the DLS to increase the ability of general practitioners who have not received additional specialist training to accurately diagnose skin conditions.”
Liu and Bui further explained that dermatologists do not only provide a single diagnosis for a skin condition, but make a ranked list of possible diagnoses (a differential diagnosis) that must later be limited by laboratory tests, imaging, procedures and consultations. This also applies to the Google researchers’ system, which contains one or more clinical images of the skin disorder and up to 45 types of metadata (for example, self-reported parts of medical history, such as age, gender, and symptoms).
Just as accurate as a dermatologist
The tests are promising and the system developed by Google AI seems to be pretty accurate. Often even as accurate as the diagnosis made by dermatologists, in the case of 26 different skin conditions. Skin cancer appears to be more difficult to detect because it is less common and the system can, therefore, be trained less well.