Interview with Hugo Aerts, professor of AI in medicine at Maastricht University (Netherlands).
He is also director of the AI in medicine (AIM) program at Harvard University and Mass General Brigham in Boston (USA).
You received a EUR 2 million European Research Council grant for the project ‘Deep Learning for Automated Quantification of Radiographic Tumor Phenotypes’ that runs from 2020 to 2025. What is the aim of the project?
Hugo Aerts: One of the greatest breakthroughs in the treatment of cancer in the past few decades is immunotherapy. Immunotherapy does not act directly on the tumour, like chemotherapy, but rather triggers the immune system to recognise and destroy cancer cells. The problem is that it is very expensive and that it only works for a subpopulation of patients. At present, we are not good at predicting for whom it will work and for whom it won’t. But for the patients for whom it works, it can prolong life from a couple of months to a couple of years. My European Research Council project aims to develop artificial intelligence (AI) techniques that can predict which patients will benefit from immunotherapy and which ones will not. We focus mainly on lung and skin cancer.
Can you describe the road to reach that goal?
Hugo Aerts: We have already published a couple of proof-of-principle studies demonstrating that our AI models can improve the prediction of the value of immunotherapy for a patient. When CT scans provide a detailed three-dimensional image of the full tumour, AI is able to quantify the information in a detailed and automatic manner. Now we want to further improve the models by collecting a large number of CT scans from hospitals in both Maastricht and Boston.
Hugo Aerts: In parallel, we are still improving our deep-learning-based AI technology. A tumour can be very heterogeneous, meaning that in some areas it can look very different from in others. Our technology can quantify the heterogeneity within the tumour better than doctors do. After testing the technology on independent data, we hope to show that there is a significant improvement in predicting the value of immunotherapy for a patient. If that is the case, then this could be the basis for starting clinical trials, hopefully by 2025.
What is your general expectation for using AI in medical imaging in the next 10 years?
Hugo Aerts: The biggest problem with many AI solutions at the moment is that, despite being very promising, they are difficult to integrate into a clinic setting. Radiologists have to perform too many steps. Therefore, we explicitly want to build our AI into the systems that radiologists and oncologists use. Within only a few minutes, they should get the result of the AI. So, the big challenge is how do we develop clinical solutions that really improve the outcomes for patients and that are minimally disruptive to the present clinical procedures?
Hugo Aerts: A second trend that I see happening is AI becoming a kind of polar star towards which all hospitals want to navigate. Sometimes there is too much hype in AI, and some disappointment will undoubtedly follow, but the good thing is that the AI excitement has finally motivated hospitals to create a 21st-century data environment where data is easily searchable.