Image recognition in combination with learning systems is already making reliable diagnoses possible. Self-service early recognition could make an important contribution to health care.
According to a Wired report , scientists at Stanford University in America have succeeded in teaching AI software to recognize skin cancer. To do so, they used a deep-learning algorithm for image recognition and trained the software using 129,450 photos of 2,032 skin ailments. On the subsequent test, the software achieved the same results as three dermatologists who were used as a control group.
The authors of the study , published in the magazine Nature, say that their AI is fully scalable, meaning that in a few years it could be used in diagnosis programs on smartphones. However, it must be mentioned that a possible limitation is presented by the fact that the photos of the skin ailments were captured in a clinical setting with special lighting. The results can therefore not be transferred to everyday situations in the doctor’s practice, at least at the moment. It will probably be possible, though, to further train the artificial intelligence with images produced by smartphone cameras.