The aim of the collaboration is to treat cancer more effectively with the aid of artificial intelligence. Prior to and during cancer treatments, a great deal of complex information about patients becomes available through medical imaging, pathology, DNA, etc. It remains challenging for medical specialists to choose and implement the best treatment based on all this information. The use of self-learning algorithms could offer a solution.
Knowledge domains combined
The collaboration combines expertise in cancer research with that in AI techniques. These knowledge domains are represented by Jan-Jakob Sonke on behalf of the NKI, specialising in image-driven therapy, and Marcel Worring on behalf of the UvA, an AI expert.
‘The first step for us will be to roll out one or more self-learning algorithms developed by the ICAI lab within the Antoni van Leeuwenhoek hospital,’ says Sonke.
‘AI technology will play an increasingly important role in scientific research into cancer,’ says Worring, professor at the UvA’s Informatics Institute. ‘Self-learning algorithms can sometimes take over tasks that normally require human intelligence. One family of self-learning algorithms is ‘deep learning’. Instead of the programmer explicitly stating which parts of the data the algorithm should focus on, he or she only specifies the end result. The advantage of this is that the deep learning algorithm can come up with insights that the programmer hadn’t even thought of.’
In the case of image recognition for the detection of cancer, a large database of medical images from patients with cancer is used. The computers learn from this and, if the database is large enough, can already recognise certain tumours as well as a specialist. AI-driven image analysis can allow for a therapy to be applied in a much more patient-specific manner.
‘There are a number of aspects that we want to realise,’ says Sonke. ‘For example, developing algorithms that can make semi-automatic treatments possible. This would allow for the flexibility and frequency at which we respond to changes in the patient to be greatly refined. In addition, in the context of personalised medicine, we want to use images to predict which patient will benefit most from which therapy.’