The stream of digital data flows, in the meantime, of course, the water from our taps. The Internet of things covers the whole world with data points. By 2020, the company, IDC, estimates the volume of digital data will reach 40 Zettabytes. This is a number with 21 zeros!

data comes from many sources and are available in different Form and quality: as pictures, texts, Videos, or audio files, or notes that need to be digitized. Personalized medicine uses a variety of data sources – from Fitness trackers on the individual history of the disease to the genetic make-up.

naked data bring anything to anyone

But the raw data is of limited value. Only when these are prepared in such a way that they can be dealt with by using Algorithms, is it for the science, the medicine, or the digital economy interesting. Similar to the conversion of raw materials to the finished product, there are also data to a multi-stage finishing process, at the end of which a new realization is.

In intensive care stations of hospitals, the patients are under increased observation. The Doctors and the care team is confronted with a flood of measurement values and information, which must be continually observed and interpreted. To find out how far artificial intelligence can support these tasks, have developed the Bern University hospital and a research team at the ETH Zurich, an early warning system. The goal was a circulatory failure of patients to the intensive care unit up to eight hours before it occurs, predict.

All 5 minutes a prediction

In a first step, it was a matter of the huge amount of data Inconsistent to filter out and make sure that you measure it in the Same way. 36’000 remained of 54’000 archived Patient records to the analysis, and from the 4500 variables finally, the 20 most relevant were determined. Based on these data, the researchers developed a model that takes all of five minutes, a prediction ends are then methods of machine learning.

One of the challenges was to define an optimal threshold value at which an Alarm is triggered. A lower threshold value of the number of increased false alarms are too high, missed a sticky situation. The researchers were able to finally prove that their model was able to say failure in 90 percent of cases, a cycle in advance, and are on average two and a half hours in advance – this in only one to two false alarms per day and Patient.

The man brings the data to Speak

conclusion: From the mountain of data to start with, it has successively developed a new understanding and a tool will be created that can help Doctors and nurses in the ICU to save lives. The research will not end there. It is planned that the early warning system in a clinical trial further testing.

Our mountain of data is growing from year to year – and thus the potential for meaningful applications of artificial intelligence in medicine. It is important to bear in mind that the data can only be by human judgment, and critical examination of the Talk brought. A wonder machine that you feed with data, and at the end of crystal-clear conclusions spits out, does not yet exist.

your Joël Mesot