Improving cancer radiotherapy with AI

How does CWI help doctors and businesses to improve fighting cancer? Professor and group leader Evolutionary Intelligence Peter Bosman (CWI and TU Delft) talks about his research on artificial intelligence (AI) for medical sciences, in which he collaborates with hospitals and companies such as Elekta.

Share this page

Peter Bosman:

Radiation therapy is one of the most important ways to treat cancer. With smart algorithms, we can give doctors a unique insight into the best treatment plans for individual patients.

Brachytherapy

Radiation is one of the main ways to treat cancer. A central problem here is that you want to hit the tumor, but not the surrounding organs. This also plays a role in brachytherapy - internal radiation, which is used, among other things, to treat prostate cancer. Making a good radiation plan must be done relatively quickly, because the examination is uncomfortable for the patient. First, catheters are inserted, through which later a radioactive source is passed. Next, an MRI scan is made to see which organs are involved, what they look like and where exactly the catheters are located. A radiation plan is then made on the basis of drawings: how long should the radioactive source stay in which place? By hand, optimizing such a plan takes quite a long time. Finally, according to the plan, a radioactive source is passed through the catheters that will provide the radiation.

Better decision making

For patients, it is not pleasant to have to wait a long time for a proper radiation plan while the catheters are in place, which is why doctors want to reduce this time to one hour. Previously, doctors could make one radiation plan in that hour — a race against time. Peter Bosman: "However, with the help of our AI, it is now possible to make a whole series of these radiation plans representing different considerations in just a few minutes. This leaves more time for qualitative decision-making by doctors, so there is more time to consider 'what is actually the best plan for this patient, how can we optimally irradiate and still save as much healthy tissue as possible?"

More on the collaboration between CWI and Amsterdam UMC

Show

Peter Bosman: “Before our method could be implemented in a hospital, it was of course important to gain support for this and overcome resistance. That's why we validated our method with physicians, and had them assess plans without telling which plan came from where: from doctors or the computer. We asked: which plan would you prefer to use for this patient right now? In 98% of the cases, they chose a plan that we had created with the computer. That was the deciding factor to start working on a version that could be clinically used.”

This method to better fight prostate cancer with AI was developed at CWI, in close collaboration with the radiotherapy department of Amsterdam UMC and Elekta, a Swedish technology company that develops radiation equipment and software for the treatment of cancer.

Peter Bosman: “Collaboration requires a lot of coordination. As a team, we worked distributed over CWI, AMC and the Elekta location in Veenendaal. In order to learn to speak each other's professional language, all researchers met twice a week (before corona times), alternately at CWI and AMC. In addition, the researchers went to Elekta every four weeks and we had an update meeting every six weeks where everyone was present, including doctors and clinical physicists. This allowed us to look at results together and learn from each other what is important and what the needs are. I recommend such intensive coordination to everyone.” This collaboration made it a success.

This research is currently being continued in the field of cervical cancer by CWI, Leiden UMC (Leiden University Medical Centre) and Amsterdam UMC, funded by KWF (Dutch Cancer Society) and Elekta.

Peter Bosman: "In addition to optimizing radiation treatment plans with AI, our group also works on solving complex image processing tasks, for example the registration of two MRI scans. This means that we want to know for each voxel in one of the MRI scans where it goes in the other scan. In case of large deformations or content mismatch, this is a very difficult, unsolved problem. We work on solutions for this based on integrating biomechanical modelling with multi-objective optimization, also in collaboration with Elekta.”

“Our primary form of AI is based on our long-standing research on evolutionary algorithms (EAs), which we started in 2004. In 2012, we applied the first concept of EAs to image registration, a technique in image processing that aligns multiple scenes into a single integrated image and prevents issues such as image rotation, scaling, and skewing that often occur when overlaying images. We use the concept of natural evolution to enhance solutions, allowing the computer to cleverly identify what properties make a solution good and exchange those properties between solutions, similar to plant breeding but for computational problems. We have methodically and technologically improved and accelerated this EA method, making it more than 100,000 times faster for problems like image registration, partially through parallelization on hardware such as GPUs (graphics cards)."

Our methods are not only effective and efficient, but also applicable to various complex problems, especially in situations where multiple conflicting goals need to be achieved. The algorithm learns itself and we can also add additional knowledge during the learning process, such as in the case of brachytherapy for prostate cancer knowledge about the build-up of radiation doses through the inserted catheters. In the end, this resulted in an algorithm that produced much better results than others.

Peter Bosman: "My ideal is that a 'plug and play' solution will be available for all hospitals and all types of radiation therapy but we are not that far yet. First we are working with different hospitals and companies on how we can improve the method even further".

To help support your medical practice in the best possible way, we at CWI design advanced methods that can make optimal use of the increasing amount of data and computing power. In doing so, we help to improve the quality of medical treatments. We do this by deploying flexible, fast and reliable methods from artificial intelligence in an interdisciplinary team. Are you a brachytherapy physician or a radiation equipment design company with questions or interested in collaborating? Please feel free to contact us.

Contact

Prof. Peter Bosman is group leader of the Evolutionary Intelligence research group at CWI and a part-time professor at Delft University of Technology.

More efficient cancer radiation therapy with help of AI (in Dutch)