To increase the chances of successful surgery for breast cancer, patients sometimes undergo chemotherapy to shrink the tumor prior to surgery. This therapy does not always work, but it does have unpleasant side effects. Researchers can use artificial intelligence to predict for whom preoperative chemotherapy is useful and for whom not. But who leaves such a drastic decision to a computer? In a large international study, Maastricht researchers are translating technology into practice in order to increase trust in artificial intelligence (AI).
Artificial intelligence is a computer system that is able to ‘learn’ from examples. For example, such a system can learn to distinguish between an apple and a pear by analyzing thousands of images of apples and pears and determining which characteristics make them different. If the system then sees a new photo of a pear, it determines which of the two types of fruit it is based on, for example, the shape, color and size.
Maastricht professor Philippe Lambin has developed a method for using artificial intelligence in medical images, such as MRI, CT or PET scans. This method is called ‘radiomics’. In the case of breast cancer, a computer system uses MRI scans to distinguish between tumors that shrink as a result of chemotherapy and tumors on which chemotherapy has no effect. In MRI scans of new patients, the system then determines whether chemotherapy will be successful, for example based on the shape and structure of the tumor tissue. If this is not the case, the patient can be spared chemotherapy and alternative treatments can be looked at more quickly.
Although the technology has now proven itself, radiomics are not yet used in healthcare. That’s not surprising, says Lambin. “You now get a decision based on artificial intelligence, but the doctor has all kinds of questions about such a decision. Why does the system choose chemotherapy and not surgery? When does the system find that chemotherapy is successful, and when is it not? Doctors will not accept such a decision without proper explanation, without questioning or checking the decision.”
At the moment it is not possible to ask the system questions, but Lambin is changing that. “A kind of dialogue with artificial intelligence”, he calls this development. “Suppose the system decides that chemotherapy is the right treatment choice. Then the doctor can, for example, request an MRI scan of a patient whose chemotherapy was successful, to see the similarities between the scans. Or they can study the differences with the MRI scan of a patient for whom chemotherapy has not helped. They can also ask the system for an estimate of the effect of alternative treatment options, or what the consequences will be if the doctor does not follow the recommendation. This allows doctors to check the decision of artificial intelligence and determine whether they agree with it. Only when doctors understand how the decision is made will they start using the technology for their patients.”
The translation of scientific technology into a useful method for doctors is part of ‘RadioVal’, a large study in which eight countries from four different continents collaborate to investigate the accuracy, safety and usefulness of radiomics for breast cancer patients. The research has not been set up worldwide for nothing. Lambin: “Radiomics must be able to deal with images from different brands of scanners, with rarer forms of breast cancer and with forms of breast cancer that are more common in certain parts of the world. The technique must be fair, and make a good prediction about the best treatment choice regardless of such differences.”
RadioVal is not the only international research building on Lambin’s work in radiomics. “Maastricht has now become the European center for radiomics,” he says.
The RadioVal project started in September.