"Data can save lives." With this announcement, Roland Eils, founding director of the "Digital Health" center at the Berlin Institute for Health Research, advertised on Thursday at a PR-related event of the Federal Press Office for better availability of research-relevant patient information. This is necessary in order to be able to use artificial intelligence (AI) concepts in healthcare on a large scale. It is therefore important to have nationwide guidelines for patient data donation.
Therapy recommendations from the computer
show more
Data-driven genome sequencing in particular now brings clear advantages in clinical care, the former head of the "Bioinformatics and Functional Genomics" department at Heidelberg University justified his plea. In a project there with 1800 seriously ill cancer patients, using the genome decoding at the end of the computer recommended therapies that "worked wonderfully". For 80 percent of the participants, the AI system used gave advice that the doctors would have implemented at least in a third of the cases. Almost half of the palliative patients responded positively to the treatment.
At the same time, Eils referred to the success of "automated support" for medical doctors in imaging procedures, particularly in the field of radiology, and thus poured water on the mills of Andreas Lemke, founder of the Berlin startup Mediaire, which works on assistance systems for analyzing medical image data in radiation medicine. "A radiologist only has three seconds to make a diagnosis," said the physicist. This is where the AI can start, and relieve the human expert of "simple tasks" in the detection of signs of Alzheimer's or multiple sclerosis, for example.
The radiologist usually receives a CD with images from the radiologist. Lemke wished that such formats should already be made available to researchers. In addition, there is already a fairly large pool of scientific data in this medical sector, which is "almost too good for forest and meadow radiology". In order to establish generalizability, "there cannot be enough data" as long as it is of high quality.
Better diagnostic systems also promote unnecessary treatments
But with some success reports from this area, caution is advised. At the beginning of the year, researchers from the USA and Great Britain made headlines with a study according to which Google's AI DeepMind could detect breast cancer better than radiologists. The error rate when evaluating mammograms was up to 9.4 percent lower.
Berlin psychologist Gerd Gigerenzer and his team from the "Unstatistics of the month" refer to itthat in the UK the AI system was on average only slightly better than the first radiologist and slightly worse than the second and the consensus judgment. The authors of the study themselves had made it clear that the results in general would probably have been better with specialized radiologists.
The "non-statisticians" point out: "The better the diagnostic systems, the more small and clinically irrelevant forms of cancer are discovered, which are only technically cancer." Since these harmless forms could not be distinguished from others at the time of early detection, many unsettled patients already received unnecessary operations and radiation or chemotherapy. The total number of women who die from cancer does not change through screening.
. (tagsToTranslate) AI (t) Algorithms (t) Artificial Intelligence (t) GDPR (t) Data Protection (t) Artificial Intelligence (t) Machine Learning (t) Missing Link (t) Regulation



