How computers support doctors
Specially trained computers can sometimes make more accurate diagnoses than doctors. They could support medical professionals in many fields
Voice assistants such as Siri or Alexa, who turn off the light in the bathroom if desired. Navigation systems that indicate the risk of congestion and suggest detour routes. Internet advertising banners that recommend additional products after making an online purchase. Artificial intelligence (AI) has long since found its way into our everyday lives. And now medicine has also discovered the potential of this technology. Almost every day, research groups report new ways in which AI could support doctors - in making diagnoses, in choosing a therapy or in monitoring the progress of an illness.
To put it very simply, artificial intelligence works like the human brain. Computers are first fed a lot of data, which they then use to recognize images or provide answers to a question. Experts speak of machine learning here, and when it is further developed, they speak of deep learning or neural networks.
Machines can recognize patterns better
But can you rely on such results? Does the computer recognize cancer in the X-ray image just as well as a doctor? Or even better? "Artificial intelligence looks much deeper into the biology of a tumor than the human brain can," says Professor Michael Forsting, chief radiologist at Essen University Hospital. "It recognizes patterns that we don't see." This is why studies have come to the conclusion, for example, that the digital doctor can differentiate between benign moles and melanomas better than most flesh-and-blood doctors.
Skin marks: benign or malignant?
Is the pigment mark harmless or is it black skin cancer, a melanoma? Dermatologists cannot always judge this with certainty by looking at them. Professor Holger Hänßle from the University Dermatology Clinic in Heidelberg has developed a system that supports doctors in their work-up.
In one study, it saw the difference better than most of 58 dermatologists. Only very experienced physicians were more accurate. The system is now used in around 40 practices. "The doctor must always have looked at the birthmark first," explains Hänßle, "only then can he switch on the artificial intelligence." The decision remains with the doctor. He can also take other information into account, such as whether the skin mark has changed or is a new one.
Many scientists are working on other applications, some of which are about to be put into practice. For example, a Forsting project: with a 95 percent probability, his system can use magnetic resonance imaging to predict whether the cervical cancer has already developed metastases. Without tissue removal and without the professional assessment of a doctor.
Less routine work
Programs by other researchers, for example, detected breast cancer on mammograms at least as often as doctors. The diagnosis of lung cancer is also one of the projects that have come a long way.
For radiologist Forsting, however, AI initially has another benefit. You can relieve the doctor of many routine tasks. For example, counting the foci of inflammation in patients with multiple sclerosis. Or the measurement of the tumor size during check-ups.
Among other things, this could prevent errors that arise from the "satisfaction of search", as doctors say. So from the satisfaction of having discovered the finding you were looking for. This then easily leads to the fact that other abnormalities are not noticed. The number of MS foci has been determined, but the metastasis in the diagnosis area is overlooked. If the computer takes over the routine, the doctor could focus on whatever else the recordings may reveal.
One doctor, 20 intensive care patients
Dr. Alexander Meyer im Sinn, prospective cardiac surgeon at the German Heart Center in Berlin. He was shaped by the experience he had to make as a young doctor: as the only doctor in an intensive care unit to look after 20 patients.
Meyer has now fed a computer with the data of 11,000 patients and developed an assistance system from it. This collects the large amount of data that is
private patients are collected, brings them together and analyzes them. "This enables the computer to identify complications at an early stage that are not yet clearly evident and can escape the doctor, especially in hectic situations," explains Meyer. The AI can even prevent deaths this way. For example because it enables faster intervention in the event of kidney failure.
Better than man?
Computers can also help diagnose cerebral hemorrhage undiscovered by doctors. You will find cardiac arrhythmias in the EKG. They have the potential to recommend therapies for blood poisoning. They recognize pulmonary embolisms.
But the AI does not work perfectly. The accuracy of their diagnoses is often over 90 percent, but almost never 100. Still, that's often enough to outperform doctors or at least be on a par with them.
The artificial doctor can learn from this - through additional data that is fed into him. Whether you can trust artificial intelligence generally depends on their quality. The amount doesn't matter. It is crucial that the computer is fed the correct information. This is shown, among other things, by the experience of researchers who had developed a system for examining the liver from computer tomographies.
It worked well - but failed organs from Asian patients. The anatomical position differs a little from that of Europeans. The program was only able to recognize the Asian livers as such after entering the relevant images.
The endurance test is often omitted
Such errors make it clear why quality control of the systems is important. It is first carried out by examining the accuracy of the computer system. Before a market launch, an endurance test in real situations would also make sense, if possible in direct comparison: medical diagnosis alone versus medical diagnosis with computer support. But because the systems are classified as medical devices, such tests are not mandatory and are often not carried out.
On the other hand, medicine now almost naturally trusts the computer in other fields as well. In the past, for example, medical employees counted the cells in a blood sample under the microscope, today automated systems do it - and do so much more reliably and with fewer misdiagnoses than trained specialists.
Help or replacement?
However, many doctors are concerned that the processes in the computer ultimately represent a kind of "black box". In which way, with which analysis sequences a diagnosis or a therapy recommendation comes about is usually unclear. Doctors don't like that any more than patients. IT specialists are therefore already working to ensure that computers provide a kind of justification with their result.
In any case, one thing is clear: no matter how much digital help will be of use in the future, it will not replace the doctor. On the one hand, the analyzes carried out by a so-called algorithm only ever apply to a specific, limited task. He cannot analyze an intestinal polyp (see box) and diagnose an inflammatory bowel disease at the same time.
Robot doctors remain a fantasy
On the other hand, AI can often recognize certain patterns more precisely than humans, but not emotions and personal backgrounds. How a patient experiences their illness, what history it has, what hereditary problems are present, what the physical examination reveals: All of this is often just as important for targeted treatment as the exact findings in a picture - which can sometimes be misleading. For example, not every herniated disc hurts, while a visually healthy back can still hurt a lot.
Experts nevertheless agree: The increasingly important role of AI in medicine harbors risks, but above all many opportunities - if people keep the scepter in their hands. Fantasies in which we will be treated by robots and computers instead of doctors in the future will therefore remain exactly that for the foreseeable future: fantasies.