Artificial Intelligence by CWI and Amsterdam UMC proposes the best radiation treatment plans in clinical practice for the first time – Centrum Wiskunde & Informatica (CWI)

Original article can be found here (source): artificial intelligence

CWI researchers, together with the department of radiation oncology of Amsterdam UMC, have developed software based on Artificial Intelligence (AI) that quickly proposes multiple radiation treatment plans for each patient. The software functions as a type of ‘route planner’ for the doctor: it presents multiple plans based on the data of the patient that represent trade-offs between giving sufficient radiation dose to the tumor with as little damage possible to the surrounding organs. This not only helps the doctors to make plans faster, it will also improve plan quality. Amsterdam UMC has treated the first patient with a plan proposed by the new AI on March 17, 2020. The innovative technique will be used for the treatment of prostate cancer with internal radiation.

Radiation demands delicate maneuvering

Radiation is one of the most important treatments for cancer. In the case of prostate cancer, brachytherapy -a form of internal radiation with the use of catheters- is a very suitable treatment. Multiple catheters are inserted into a patient, through which a radio-active source is being led. After insertion of the catheters, a treatment plan is made by the doctors. In the case of brachytherapy this is being done while the patient is waiting with the catheters that have been inserted earlier. An uncomfortable situation, that preferably lasts as short as possible. Usually it takes the doctors quite some time to design a treatment plan for a specific patient. It takes the new software only a few minutes to come up with a whole range of treatment plans, however. Each plan states for how long the radioactive source should be focused on a specific area, in order to deliver a specific amount of radiation dose in that spot. The focus is on establishing a treatment that gives the desired amount of radiation dose to the tumor, while sparing as much of the surrounding healthy tissue as possible. 

In the research and development stages, a blind test was performed with a team of radiation oncologists, showing both plans previously made and used in the clinic and new AI-made plans for the same patients. The radiation oncologists were very much convinced by the abilities of the new AI technique: they preferred an AI-based plan in 98% of the cases.

Peter Bosman, senior researcher at CWI’s Life Sciences and Health group and project leader: “Our form of AI delivers a spectrum of plans very fast, that represents the trade-offs between delivering sufficient radiation dose with as little damage to surrounding tissue as possible.  This gives immediate insight into what is feasible for a specific patient. This relieves doctors from undertaking a complex approach to configuring a treatment plan using existing software that requires intensive human-computer interaction.”

“Amsterdam UMC has performed research in the area of computer-aided support for making radiation treatment plans for years already,” says Arjen Bel, head of the clinical physics of the department of radiation oncology. “The challenge is to make these plans quickly, as well as of high quality.”

Bradley Pieters, radiation oncologist at Amsterdam UMC: “The improved radiation treatment plans can lead to better results for patients with prostate cancer. The extra time allows us to deliver tailor-made plans for the patient. Besides that, we can now use our medical knowledge optimally, as well as extra knowledge about the patient that the computer does not have.”

Unique collaboration

The development of the new software was made possible by a close research collaboration between CWI’s Life Sciences and Health group, the department radiation oncology of Amsterdam UMC-location AMC, and Elekta, a company that delivers radiation equipment and software to hospitals. The team decided to develop software for this problem with a form of AI (evolutionary algorithms) at its core. These algorithms are very suitable to effectively and efficiently search for good solutions to complex problems, especially when there are multiple conflicting goals to be achieved.

The team especially focused on a form of evolutionary algorithms that arise from Bosman’s long-running research line. These algorithms display intelligent search behavior. They have the ability to analyze a certain problem and subsequently teach themselves how they can come up with better solutions for that problem. The research team made special adjustments to search for configurations of treatment plans in the case of brachytherapy for prostate cancer as good as possible. They did that by letting the algorithm use knowledge about the build-up of radiation dose from the inserted catheters. Ultimately, much better results could be booked with this algorithm than with other algorithms.

Broadening the scope

One great advantage of the developed AI-software is that it can be expanded to other types of cancer relatively easily. Calculating a spectrum of possible treatment plans, considering the trade-off between a good chance of a successful treatment and possible side effects, is something that is needed in the making of radiation treatment plans for multiple forms of cancer. A follow-up project has already been planned. With funding of the Dutch Cancer Society (KWF Kankerbestrijding), CWI, Amsterdam UMC, and Elekta will expand this research to the area of internal radiation for cervical cancer, led by new project partner Leiden University Medical Center (LUMC).

Tanja Alderliesten, senior researcher at LUMC (previously Amsterdam UMC) and project leader: “This time we will even execute a national validation study in order to create national impact.” Business partner Elekta is working on the worldwide distribution of the software, to make sure the rest of the medical world can also profit from this innovation.

YouTube video about the project (in Dutch)

Interview with Peter Bosman in Bits&Chips (in Dutch)