How AI can improve healthcare and medicine

Artificial intelligence (AI) will make medicine the best medicine ever. It will save millions of people suffering and make health systems around the world fairer, more humane, more efficient and safer. Sound exaggerated? But it is - very likely - not.

Medical AI systems are another technology that helps humans push the boundaries of their body and mind. He invented the plow because he cannot dig up large fields with his hand; the car because he doesn't get very far on foot; the electric generator because it cannot generate electricity on its own. Now the human being is faced with the task of searching the billions of health values ​​that occur today at the doctor and in the clinic for relevant patterns. He won't be able to do it with his brain alone. Anyone who has ever tried to organize a loose collection of numbers in an Excel file according to logic knows that it is extremely tedious, almost impossible. How about if a machine took over this task?

AI systems have long been in the process of analyzing the gigantic data sets that are sent around the globe every second - and quite a few of them come from medicine. It is therefore no longer the question of whether artificial intelligence will continue to change the world and thus also medicine - but only when and how.

The assessment of some critics that the technology is just hype is a dangerous misjudgment. Anyone who refuses to use new medical data is robbing the health system of its future. Who wants that?

How good it would be if the AI ​​did a lot of the medical assembly line work!

The skepticism towards artificial intelligence is often due to the idea that it is supposedly trying to imitate humans and ultimately replace them. That sounds creepy or pretty impossible, depending on your perspective. In fact, this fear is unfounded, because self-learning algorithmic systems, which are actually at issue, are not intelligent in the human sense. They have no gut feeling, they cannot easily transfer what they have learned to other problems, they have no consciousness and are certainly not able to question themselves.

Roughly simplified, self-learning algorithmic systems do nothing other than search independently in a huge pile of data for patterns that humans would not recognize or would only recognize with great effort. This is why such systems are particularly suitable for repetitive tasks such as searching for abnormalities, deviations or similarities. For example, in computed tomography scans that doctors have hundreds of times done in hospitals every day.

That brings enormous advantages. First, the systems can recognize relevant patterns that humans may never have looked for at all. In addition, unlike radiologists, they never get tired or frustrated. You work without a break and can theoretically be fed an unlimited amount of further data: You learn, learn, learn. AI algorithms relieve people of precisely those activities that dehumanize them.

Numerous studies impressively prove that AI systems are very good at this and - for example - can reliably evaluate images. For example, a sensational study from Stanford, published in the specialist magazine, showed Naturethat self-learning algorithms classify skin cancer as competently or even better than dermatologists.

Even if the results of the study cannot be transferred to everyday medical practice without restrictions, they nonetheless show the enormous potential of the technology. It is often general practitioners who have to assess a conspicuous area of ​​skin such as a mole. As in many other areas, AI systems offer good support here. Another in Nature The study that has been published shows that the combination of algorithm and doctor has a particularly low error rate, for example when assessing MRI images. Human-in-the-loop is the technical term for this collaboration. It will soon be the gold standard of modern medicine. AI systems will not replace doctors, but relieve them so that they have more time for what patients often miss today: conversations, empathy, humanity.

Medicine today is often assembly-line work, which is one of the reasons why it is met with so much suspicion. What an improvement it would be if algorithmic systems did some of this work! Increasing efficiency does not mean turbo-capitalism, but supports low-error and human health care. People want to talk to flesh-and-blood doctors, but they also want them to make decisions for them well-rested and up to date with the latest research - this is possible thanks to AI.

The point is to treat people based on the information that is already available around the world and, ideally, to heal them - actually a matter of course. But still, it is hard to believe, in hospitals patient data such as body temperature or blood pressure are documented with pencil and eraser, a procedure with an extremely high risk of errors - did the patient now have 38.7 or 37.8 degrees?

These data are at the same time a treasure trove; it is data that slumber unlinked in paper files, but from which millions of patients could benefit. Algorithmic systems are able to put these values ​​in relation to one another and recognize trends. Already today they warn in some clinics that a patient's condition could worsen - even before it actually happens. AI systems can even determine biomarkers for individual diseases or predict therapeutic success. However, the decision as to how the patient is treated is still made by a person.

This form of technical innovation is not the PR of greedy private clinics, but the basis of preventive medicine of the future that can save a lot of suffering. This also includes curbing unnecessary treatments. A study in the specialist magazine shows Radiologythat with the help of algorithmic systems in breast cancer diagnosis, the number of operations could be reduced by up to 30 percent.

The systems also help to identify extremely rare diseases that otherwise only doctors with great experience or specialization can diagnose. It doesn't matter where in the world the data is used. This enabled a new form of global health justice. AI systems can already evaluate findings today, for example to support helpers in remote areas of India or Africa in deciding which of the many patients should urgently consult a doctor - and which not.

In Germany, too, algorithmic systems can offer modern medicine outside of clinics. For example, they check the blood pressure and ECG data of heart disease patients and report any abnormalities to the attending physician. This enables close monitoring and also saves the patient many visits to the doctor, which can be tedious for them.

Independent experts must always be able to understand how the algorithms decide

In order for such applications to really gain acceptance, it is important to take the criticism of the systems seriously. A current meta-study, published in the specialist magazine The Lancet, For example, less than one percent of more than 20,000 AI studies on medical image diagnostics found trustworthy - good food for skeptics. In fact, it shows above all that better studies are needed to test the clinical suitability of the systems in everyday life. At the same time, this is an answer to the accusation that data-driven research relies too much on correlations, rather than on causalities. Thanks to AI systems, for example, new hypotheses about drug side effects can be specifically created that had not been thought of before. A specific study must then always provide final clarity.

Those who do not want to take these steps leave the field to private players, who will soon no longer be catchable. But public health care must not become dependent on tech giants like Google. It is therefore imperative that the state invest money in research and make university hospitals into places of the future that are so good that Silicon Valley programmers do not laugh at them as they have before, but want to work there.

This would also be an answer to two very legitimate concerns about the AI ​​revolution: First, it is fundamentally important that patient data is protected in order to avoid misuse. It would be a fiasco if AI systems helped discriminate against certain patient groups. A study inSciencehow algorithms systematically disadvantaged black people in US hospitals and how they actually received poorer care.

Second, you need an insight into the architecture of the systems. Algorithms also make mistakes and can produce incorrect results based on incorrect assumptions. It must therefore be possible for independent experts to understand at all times the criteria according to which algorithms make decisions. These logics must of course meet ethical and legal requirements.

But: These restrictions are not arguments per se against the use of AI systems in medicine. Because it is not technology that drives abuse, but people.

Blind enthusiasm is just as out of place as pure distrust. In order to guarantee optimal use in medicine, it must be clear that AI systems do not have superpowers that will soon cure cancer and MS. And yet they are valuable tools that make diagnostic puzzle work easier for doctors and relieve them in everyday life. As mentioned at the beginning, you can contribute to the fundamental change in medicine towards a fairer, more humane one, in short: to better medicine worldwide.