AI in Medical Physics – A Bit Like "Make a Wish"
No, artificial intelligence (AI) is not just about chatbots. AI can also help in medical physics to optimize processes and relieve the workload on specialists. Dr. Daniela Eulenstein, Research Scientist at PTW, explains what this looks like in radiotherapy and dosimetry.
What are the benefits of AI-based data processing?
The procedure was as follows without AI: A very specific problem had to be solved, and an algorithm was developed for it. This problem was then solved exactly with the corresponding boundary conditions, i.e., taking a proactive approach. With AI, this principle is reversed – it's a bit like "make a wish". The developers think about what the application should be able to do. Then they give examples to AI. This means that a large amount of training data is required to train AI. As medical physicists, we then no longer have to consider in detail how exactly AI gets the result; AI independently recognizes the underlying relationships. To put it briefly: I now have the opportunity to train a neural network according to my own wishes.
How does it work in medical physics? Do you have an example?
Medical physics can definitely call itself a pioneer in medicine when it comes to the use of AI. A good example from radiotherapy is automatic contouring. The use of AI-based contouring was introduced quite early and accepted quickly. Contouring is an essential part of the radiotherapy treatment planning process, because we need to know exactly which structure of the body is where during radiation treatments. It is therefore a matter of accurately delineating a patient's organs on a CT scan: kidney, liver, bone, etc. So far, this has been done by humans. But the more complex the structures are – think of the brain, for example, where the course of the optic nerve must be precisely delineated, since it is very sensitive to radiation – the more time-consuming the task becomes for the contouring person.
The fact that this task is taken over by AI is already commonplace in many hospitals. And employees are pleased that AI is relieving them of these monotonous and time-consuming tasks. However, the workload is not reduced to zero, but to a few percent. Because it is indispensable that a doctor or a medical physicist looks at the result in the end as the final decision-maker. Consequently, AI does not replace specialists in medical physics, but makes them more productive.
AI is also already integrated in the radiation delivery systems themselves, enabling adaptive radiotherapy. This means that we are no longer working with a treatment plan that has been created once, as has been the case so far, but rather with ad hoc planning. The patient is placed in the treatment unit, is scanned, and then the patient's individual anatomy at that moment is taken as the basis for treatment planning. The treatment plan adapts to a filled bladder or a displaced prostate, every day anew. However, the presence of a medical physicist and a doctor is absolutely necessary here as well. They check every step and must ultimately approve the AI-based planning before the patient is irradiated.