Artificial intelligence: Doctor AI - will you take over?

The head of the Institute „AI for Health“ (Artificial Intelligence for Health), Dr Carsten Marr, is sitting at the computer in his office at Helmholtz Munich. On his screen: Images of blood smears and programming codes. They form the basis for an AI model that is supposed to detect blood cancer better and faster than humans. The aim is to use AI to simplify the labour-intensive and time-consuming microscopic examination of blood and bone marrow cells. If blood cancer is suspected, specialists examine hundreds of cells from a blood smear under a microscope and categorise them. This takes hours. Will AI soon be able to do it in seconds? The high-performance computers at Helmholtz are currently being fed with over 200,000 images of cells from over 1,000 patients with various blood diseases.

But what exactly is AI, or artificial intelligence? The AI system ChatGPT itself says: „Artificial intelligence refers to the field of computer science that deals with the development of machines that can perform tasks that normally require human intelligence.“ Sounds conclusive. Applied to medicine, it means that AI generates artificial knowledge based on large, medically validated data sets. For example, these are labelled images of red and white blood cells with information about the respective cell type. The software then learns to distinguish optimally between the cell types. The aim is for the AI to recognise in a completely new image whether it is a red or white blood cell, a benign or malignant cell," says expert Marr.

Through this type of learning, computers should do the same thing that our brain does automatically, namely recognise things. How do you get it to do this? With the help of images and codes, for example. The programmers use these to teach the computers what they should recognise. Using step-by-step instructions, known as algorithms, the systems then learn to recognise patterns themselves without anyone telling them what to look out for. These patterns ultimately help the AI to recognise what an image shows. If this learning of the algorithm is repeated with the help of huge amounts of data, the patterns become clearer for computers and they recognise the images better and better.

AI: What is the current status?

Radiologist Prof Mike Notohamiprodjo knows just how helpful artificial intelligence already is in practice. He is the managing partner of a radiological practice network in the München metropolitan region. The network of doctors has been working with the development departments of technology companies for years and is trialling new solutions in the field of AI in order to permanently integrate suitable systems into everyday practice.

„What AI can now do very well is answer simple yes-no questions“, says the expert. For example, whether a patient has a tumour or a broken bone - or not. In these cases, the AI now achieves similar sensitivities to a radiologist, which means that the software only overlooks a few sick people. „The radiologist and the AI together are a good team – better than the AI alone and also better than the doctor alone“, says Notohamiprodjo.

What advantages does AI offer?

The advantages of AI are also very practical. „It does not get tired, is not hungry“, says the physician. „And it is often better than inexperienced doctors because it draws on the knowledge it has learnt from large data sets.“ However, AI should not be trusted blindly: „We are still at the very beginning and have a long development process ahead of us“, says Notohamiprodjo. He draws an easy-to-understand comparison: the development of parking aids in cars. „In the beginning, we only had beepers that warned us when we were driving too close to an obstacle. Then came the reversing cameras. Now cars can park automatically at the touch of a button.

However, the responsibility remains with the driver, who can take the wheel or step on the brakes at any time.

The same applies to medicine. but control and responsibility lie with the doctor. For example, AI is currently still reaching its limits when different diagnoses are possible, such as for a person with knee pain. There are numerous possible causes: Meniscus, Bänder and so on. KI – is still – üoverburdened with this.

How does AI help in therapy?

Artificial intelligence can be helpful not only in diagnostics, but also in therapy, for example for type 1 diabetes. Sufferers have to inject insulin regularly to lower their blood sugar levels. However, finding the right dose can be challenging. Leipzig-based start-up founder Thomas Wuttke knows this from his own experience, as he is affected himself. He used to inject to the best of his knowledge and feeling. As a result, the insulin dose was often not accurate. Carbohydrates from food and drink and physical activity influence blood sugar. So do fatigue, time of day, illness, stress and time differences. This does not make diabetes therapy easy. The consequences of incorrect dosages are hypoglycaemia or hypoglycaemia, which can be detrimental to health in the long term.

Wuttke's goal when he and his team developed a smart app for smartphones was to use digital technology to avoid such treatment errors. Thanks to AI, it continuously displays the currently required insulin dose. In addition to the app, you also need a continuously measuring glucose sensor on your body and a commercially available pen to inject the short-acting insulin.

The app wirelessly records the blood glucose values from the sensor. The carbohydrates of each meal must be entered manually. The AI observes the user's metabolism. As a result, it gets to know the diet and behaviour better every day and calculates the required amount of insulin. „ÄIf the blood sugar changes, for example due to an illness, the AI automatically adjusts the insulin dose after a short learning phase. This is new compared to other diabetes apps“, says Wuttke. The longer you use the app, the more precise the advice becomes.

What is being researched?

The users themselves decide whether to take the suggested amount of insulin when they enter the dose into the pen. The start-up from Biocity Leipzig is working together with electrical engineer Dr René Richter from the Technical University of Dresden. He has developed a smart insulin pen that registers the actual dose injected and transmits the value to the app via Bluetooth. In addition, the high-tech pen also checks the temperature of the insulin via temperature sensors. „If a critical temperature is reached, for example below zero or above 40 degrees, the app warns its users“, explains Richter. The insulin pen is expected to be launched on the market in around two years.

One ­vision: using AI to predict the individual risk of diseases such as cancer long before they break out. Another area of research: personalised medicine. In future, computers could contribute even more to creating treatment plans that are optimally tailored to each individual. „This will be possible if artificial intelligence has learnt from millions of people with a similar medical history. Based on this knowledge, the AI can work together with the doctor to find exactly the right therapy that will lead to the fastest possible recovery," says researcher Narges Ahmidi.