Should I be a pathologist?

And even after further training, it will still take years for the specialist to fill his "internal database", as the resident pathologist Griewank calls it, well. "It takes a pathologist many years or even decades to be really good in his field."

This "internal database", ie a good deal of experience, is now to be built up significantly faster using new technology. Pathologists can already send digital images of their samples to another institute in order to receive a second opinion. And intelligent image processing programs take on time-consuming and complex tasks such as counting cells. "This is a good example of a tried and tested function of intelligent image processing," says Stephan Wienert, tumor researcher and software developer. Wienert has specialized in pathology and is working on solutions such as the Ki67 algorithm for it. In a matter of seconds, this counts all cells in the specimen scan that are currently dividing. This determines the rate at which a tumor grows. Doctors need this value to select the right therapy.

Millions of records would be needed. But only a few pathological institutes work digitally

Little of this has so far been implemented in practice. Doctors like Klaus Griewank hesitate; They are held back by high costs, a lack of standards, outdated laws and billing models from the Association of Statutory Health Insurance Physicians. For example, the pathologist can only invoice his findings if he has also created the corresponding preparation. Cooperation between various experts and their computer systems is not yet planned.

That could change. Tech companies such as Google are constantly driving the development of systems that will change the everyday life of pathologists with the help of artificial intelligence.

Studies show that AI analyzes, for example of breast tissue, are just as good or in some cases even better than findings made by doctors. The great advantage of the systems: They work quickly and never get tired. In contrast to humans, they can also scan large areas of tissue for abnormalities around the clock. Other institutions, such as the National Center for Tumor Diseases, developed algorithms that also achieved better results in skin cancer diagnosis than the human specialists.

However, inadequate data sets still prevent broad application in practice. Because in order to be able to train an AI system, it has to be fed with millions of data records. But since only a few pathological institutes work digitally, this material is simply not available. In addition, the preparations each bear the signature of their laboratory and differ from one another in terms of cut and color. "In addition, human tissue, both healthy and diseased, has a strong individual character," says AI expert Wienert.

Companies like Google's parent company Alphabet could solve such problems in the future by buying data sets from around the world or one day even collecting them themselves. How extensively AI will be used in pathology from now on also depends on the legal framework. For many pathologists, AI support would be very welcome, today rather than tomorrow.

Or support at all: When Klaus Griewank closes his practice in Nieder-Olm in the evening, it is clear that it will, yes, it has to reopen the next day. Because pathology is a secondment business, practices never send samples to him again once they have been forced to switch their systems to another laboratory. And besides, nobody waits for a possible cancer diagnosis until a doctor comes home from vacation.

Griewank's solution if he should ever fail: his predecessor took over the practice and laboratory for a few days. The only problem is: he is now 70 years old. If he stops, Griewank has no one left to represent him.