Artificial intelligence (ai) is making strong headway in more and more medical fields. Doctors in UZ Leuven have now turned to ai in the treatment of head- and neck cancers.
Instead of solely having to rely on human expertise in the analysis of CT scans, ai can do the job much quicker and much more accurate. Thanks to the ai-aided computer model, doctors can come up with a much more consistent radiation treatment plan, and have more time to interact with their patients.
Head- and neck cancer is a complex type of cancer, that calls for a very precise radiation in order to prevent damage to surrounding tissue, such as the spinal cord and saliva glands. CT imagery is used to precisely mark the areas that should or should not be radiated.
This marking used to be done manually by radio-oncologists, which involved a time-consuming process, and with varying outcomes, since each specialist has his or her own way of doing things. A test, in which all Belgian radiotherapy wards were asked to mark the high risk-organs using the same set of images resulted in a wide array of results, even though all specialists followed the same guidelines.
Professor dr. Sandra Nuyts, a radio-oncologist with UZ Leuven, approached medical physicists to find ways to streamline this process, and joined forces with professor Frederik Maes of the Medical Imaging Research Center (MIRC) of KU Leuven. They devised a new tool, based on deep learning, a type of ai that is very useful when it comes to analyzing images and finding patterns in them. After testing the algorithm in a laboratory setting, the system was put to use in a clinical environment.
In a test setting, two experienced radio-oncologists assessed a set of CT-scans manually, and also assessed a set of scans that had already been examined by the algorithm, correcting it where necessary. It turned out that the second approach was 33 percent less time-consuming and also much more precise.
Presently, the ai-system is used on a daily basis in the UZ Leuven. Human experts still check the results of the alogorithm’s findings, but have yet to see major mistakes. Presently, the developers are investigating whether the ai-system could be useful in the treatment of other cancers.