Digital Health

How biomarkers are revolutionising disease research

Blood pressure, sweat, voice - we humans produce lots of so-called biomarkers, hundreds of them every second. And since time immemorial, the medical profession has used these to treat us. But it is only with the possibilities of artificial intelligence (AI) that we are finding out connections in the combination of biomarkers that a human alone would never come up with. With computer-generated biomarkers, we have the chance to discover completely new correlations for certain diseases - even before they have broken out.

We run. We eat. We sleep. We get sick. Everything that happens to our body can be translated into countless data. We can measure the sugar level of our blood, the speed of neurotransmitters, the temperature of our body, the chemical composition of our sweat or urine, the pitch of our voice, the regularity of our steps, the size and shape of our organs.

A flood of data comes together from which, for a long time, doctors were only able to determine any correlations that would be helpful in curing an illness through painstaking combination work after time-consuming observation. The digitalisation of medicine now provides doctors with artificial intelligence glasses in the data jungle, through which they can suddenly recognise connections that were previously unfathomable to them.

Computer-generated characteristics for a disease

The revolution is made possible by the mass of data. Jörg Goldhahn, project leader of the Bachelor of Human Medicine at the Swiss Federal Institute of Technology Zurich (ETH) gives an example. In the past, a doctor would order a patient with suspected Parkinson's disease into the doctor's office and observe him or her walking for a few minutes to detect gait irregularities. Today, a patient's gait and any irregularities in their movements can be monitored day and night with the help of a smartphone with a suitable programme - every second and, if necessary, for years.

With the help of the flood of data, a complex picture can be drawn that is revolutionising diagnosis, but also research into the disease, says Goldhahn. “That is the power of huge amounts of data: They are uncomplicated and continuously available, and artificial intelligence can be used to draw conclusions from them quickly.” Digital biomarkers are the name given to these collected characteristics, which are collected by computer and can identify a specific disease.

But the matter is not entirely simple. The flood of data that a human body produces is so complex and multi-layered that a human being can hardly keep track of it, let alone find the right correlations, says Joachim Buhmann, professor at the Institute for Machine Learning at ETH Zurich. That is why those who deal with digital biomarkers, i.e. who want to pull condensed information for a certain disease pattern out of the data thicket, must quickly enlist the help of artificial intelligence.

“It is unlikely that a human being can read out the right combinations from the flood of data. But a machine can,” says Buhmann. Buhmann considers the human claim to be able to understand all the correlations that a self-learning artificial system finds out to be “hubris”. “We can only control the results of machine learning by testing whether the predictions of the machine prove true in investigations,” says Buhmann.

Radiomics - several hundred computerised measured variables

Until now, the translation of imaging procedures such as radiology into data sets has been a particularly great challenge. Now engineers are announcing revolutionary news. Radiomics is the name of the digital superego that is supposed to show the image interpreter the way far beyond the limits of his previous work. An artificial intelligence is not only to collect data sets, but also to extract quantitative image parameters from the image information in order to evaluate them for automated analysis.

The radiologist, who until now has only been able to judge with the help of his personal experience, is thus given a computerised pair of glasses in his hand. And that is nothing less than a quantum leap. The computer not only measures the volume, shape, homogeneity, boundary and neighbourhood relations of a tumour, for example, just as the analogue evaluating doctor does, but also carries out analyses of the image texture and functional parameters, for example blood circulation.

Several hundred such computerised measurement parameters have been developed and tested so far and have proven to be usable and reproducible in principle.


This data can be brought together with information from oncologists, pathologists, laboratory physicians, pharmacologists and geneticists. Ever new image parameters, ever new and ever more comprehensive correlations, constant learning ability, always in coordination with biobanks around the globe will emerge.

New correlations result in new therapeutic options and thus benefit humans in the fight against diseases. The flood of data helps, for example, in the development of new target therapies. If artificial intelligence finds out, for example, that certain proteins are the main cause of malignant behaviour in cancer cells in an individual case, these proteins can be targeted with drugs.

In people who respond to treatment, it is also possible to use biomarkers to monitor the effects of therapy. If the body releases certain molecules again, this can be a sign that the cancer is returning.

The marathon runner's hip fracture was a long time coming

Those looking for a simpler example of the utility of digital biomarkers can also listen to Goldhahn's experience in the orthopaedic field: “There are studies with runners, their movement data is permanently measured. After a hip fracture, we can see after evaluating all this collected data: This injury has been announced. For example, the gait became more unstable and unsteady beforehand. In this respect, the flood of data can be used preventively in the future. So a simple chatbot could warn the walker about the injury and give instructions for action.”

Some researchers are already saying that doctors would become a superfluous species as a result of intelligent machines. Goldhahn does not want to go that far. But at least he expects the digitalisation of biomarkers to lead to a new “maturity of the patient”. People can collect a lot of information themselves through so-called wearables and have it translated into therapy suggestions through the use of artificial intelligence.

Mature patients, superfluous doctors?

“Quite banally, for example, the built-in algorithm of my smartwatch tells me that I've been lying on the sofa for three hours now and that I finally need to move again,” says Goldhahn. It can work the same way when it comes to administering medication. In addition, risks could be recognised more quickly, and therapy could begin before the disease causes irreversible damage.

A stay in hospital could lead to computer-assisted rehabilitation in the patient's own home after a few days. “The patient would then be more independent of the hospital, but still well monitored,” says Goldhahn. And the doctor: would be relieved of time-consuming follow-up examinations and could invest more time in things that the machine is not so good at: listening, being empathetic.

A lot of research is needed to get reproducible biomarkers

The crux of relying on artificial intelligence, Goldhahn believes, is validability. “A lot of research is needed to design digital biomarkers so that they can be reproduced at any time. Only then can we align therapies to them. If the pedometer can be tricked by me moving it back and forth while sitting on the sofa, the result is lack of exercise.

That's not so bad at first, but if the administration of insulin depends on the digitally generated biomarker, then such an error can kill a person,” says Goldhahn. So the data collection and the hypotheses of the machines must be followed by a whole lot of testing. The question to be asked: Does the computer's assumption prove true in every case?

And what about data protection? Joachim Buhmann calls for a paradigm shift in the way this issue is dealt with, which is characterised by timidity. Certainly, data on behaviour, personality profiles or sexual orientation are sensitive. But in his opinion, genomic and biomedical data are different.

“The problem is not that someone has a flood of biomedical data about my body. The problem is that health insurance premiums increase when a disease pattern becomes apparent. Wouldn't it make more sense to reorganise the premium system? Then we could keep our freedoms and at the same time enjoy the benefits of progress.”

Text: Ronald Voigt

Most popular