Contents page 1 — Dr. KI will see you now-page 2 — “Everything is only shown as the big achievement” On a page
Our brain is built around pattern recognition. A broad nose, long eyelashes and thin lips, it is composed of, for example, to a face, the twelve Doric columns and the Quadriga on the roof of the Brandenburg gate. To recognize patterns also Doctors to learn. In the hospital, the trained ability to save lives, if, for example, a skin doctor realizes that an alleged Mole is skin cancer. Or if a doctor discovered a tiny spot in the x-ray image and his patient to a professional for transfer.
the brain is usually very accurate. Again and again, but under a mistake of him and the doctor so in the worst case, a misdiagnosis. In addition, the Doctor often takes a long time to recordings and evaluate them. Because image recognition software has made in the past few years, significant progress, to assist Doctors in their work. The potential, as experts agree, is huge: more accurate diagnoses, shorter wait times, maybe even more time for the doctor-patient conversation. With the hopes hand-in-hand, the concern is: Is lying to the Software, some radio and some of the skin doctor replace?
image recognition software is typically used by a Form of artificial intelligence, Deep Learning. A person of the Software is a picture and a particular characteristic that needs to know it in other recordings. To find the predetermined characteristic, divides the images into several levels to get to individual pixels – the bottom layer – to form a whole picture – the highest level. Our brain makes it similar, if it must assign certain things we see, hear, or feel. Due to this analogy, the programs are also referred to as an artificial neural network. In medicine, the Software learns, by to a images – such as CT-Scans or photos of suspect skin – to- read, and on the other hand, the advance of a doctor’s specific diagnosis.
No skin doctor diagnosed only on the basis of photos
The most recent example of such a self-learning system could improve the medicine, comes from Heidelberg. Developed by scientists is a Software to distinguish the dangerous black skin cancer, malignant melanoma from harmless moles. The Software is trained with 100,000 photos, and the correct diagnosis. Then you let the Software against skin doctors to compete.
58 dermatologists from a total of 17 countries, with different experience evaluated the same 100 pictures of the machine. The pictures were of course not new, neither the machine nor the Doctors knew who she was, or the correct diagnosis. On 80 of these pictures of birthmarks were seen, and on the remaining 20 melanomas.
the results of The competition have appeared Recently in the journal Annals of Oncology (Haenssle et al. 2018). They are – at least at first sight – very impressive. The Computer hit 95 percent of skin cancer cases, indeed Alarm, at the Doctors, up from 86.6 percent. In a second attempt, the Doctors were given additional information, such as age or gender to the photos. Thus, they came to 88.9 per cent correct positive Tests. A small improvement, but still worse than the Software.
In a very important point, however, the Doctors suggested the Software. The Computer recognized the Healthy not as accurate as the experts. So he simply held 63.8 per cent of the Healthy for being healthy, the Doctors at least 71.3 percent. In other words, The Computer gave 29 patients with a skin cancer diagnosis, although the spots on the skin only mom were male. The Doctors were, however, using the additional information, only 20 false positive diagnoses.
What is not taken into account, the study also: Not a skin doctor makes the diagnosis only on the basis of a photo. In the normal case, it performs a full investigation, he interviewed the patient according to their medical history and risk factors, for example, frequent Sunburns in Childhood. In addition, he touched the spot carefully, to find out if the place is hardened or bleed easily starts, both possible indications of skin cancer.
Also reports the Software only, whether or not a Patient black skin could have cancer or not, but not whether it is in the latter case, a completely normal mole or white skin cancer that occurs in Germany, almost six times as common as malignant melanoma.