The deployment of artificial intelligence could lead to substantial advances in cancer care, but collaborative efforts are needed to increase trust among end users
The application of artificial intelligence (AI) in cancer care is hailed with a cautious optimism by the oncology community, after a number of AI tools, mostly for radiological and pathological image analysis, were approved in the USA and the European Union (EU) recently (NPJ Prec Oncol. 2024;8:72). According to the experts from various fields in cancer research and industry, as well as in regulatory and patient advocacy gathered at the 2024 RoadMAP to the Clinic - a day featuring lectures and panel discussions around AI promises and challenges during the Molecular Analysis for Precision Oncology Congress last October -, the field is entering a new exciting era where collaborative efforts and education will be key to overcome the current hurdles to AI implementation in clinical practice. Dr Mireia Crispin Ortuzar from the University of Cambridge, UK, who will co-chair the first edition of the ESMO AI & Digital Oncology Congress (Berlin, 12-14 November 2025), explains how the AI field applied to oncology is evolving.
What are the emerging fields of application of AI tools in oncology?
At the moment, the majority of the regulated AI medical devices are for radiology, but there are other fields that have been gaining momentum recently. An example is AI-powered computational pathology, for which there are also a number of approved devices on the market. Computational pathology provides the perfect ground for the deployment of a new generation of deep learning models called foundation models. They are large deep learning neural networks that learn from huge datasets in an unfiltered manner, allowing us to detect much more subtle signals in the data (Cancer Cell. 2024 Feb 12;42(2):209-224.e9). It is a very rapidly evolving area which has a wide applicability in oncology because of the availability of tissue samples from the majority of patients at some point during their treatment course, whether from a biopsy or a surgical specimen.
The other very exciting area is that of language models, which can process large and diverse volumes of text from, for instance, electronic health records to mine patient journeys and understand the evolution of the disease over time. Another emerging field is the move from a static model to a more dynamic scenario led by AI agents which can act autonomously, and dynamically call other more specific AI models to perform specific functions. Even though they are in their very infancy, there is a lot of excitement around the role they could play in data integration scenarios.
Does the recent approval of AI tools from regulatory agencies reflect a regulatory framework which is ready to understand and catch all the promises from AI?
All of the regulatory bodies and policymakers have been working hard to ensure that they are staying up to date with research advances. In Europe, the EU AI Act, which entered into force last summer, is the first-ever legal framework on AI addressing the risks in the field, and a similar approach is being discussed in the US too. AI is here to stay and is progressing really quickly, so it is likely that regulation will equally need to keep evolving and adapting. Currently, regulatory approval and the capacity of regulatory frameworks can be a limiting step to a quick adoption of AI-based cancer therapies, however there is a lot of willingness to work together as we witnessed during the final panel discussion at MAP 2024 which involved experts in regulation and clinical trials.
Of course, any change takes time, and there are still some hurdles to overcome to take a step forward and make AI a consolidated reality in oncology, not only from a regulatory perspective, but also from the point of view of data availability and scalability of findings from small studies. Progress is slower in oncology than in other fields using AI, but that is not necessarily a bad thing: we definitely do not want to “move fast and break things” in healthcare.
Despite some oncology professionals are optimistic toward AI application in cancer care, some others still show some reluctancy. What can be done to further increase trust in AI in the oncology community?
What the literature shows is that there is a significant correlation between how physicians feel about AI and how much they know about AI systems, coding, programming, etc. (Sci Rep. 2021 Mar 4;11(1):5193). It is not surprising: if doctors have never interacted with AI before, why would they trust its outputs? I am a particle physicist by training –which is often perceived to be a scary subject –, and I know from experience that once you have a good teacher who explains concepts in an easy and engaging manner, physics becomes simple and exciting. Education and training are crucial to increase trust in AI in medicine. However, every professional, particularly in oncology, is extremely busy, overloaded, and has very limited capacity to expand their knowledge and skills beyond what the daily work routine requires. As a community, we need to demystify AI and provide loads of opportunities for oncologists to familiarise themselves with the basic ideas behind it. The ESMO AI & Digital Oncology Congress 2025 represents a proactive step undertaken by ESMO toward this direction. All the congress sessions will be education-oriented and cover different areas of development and application of AI within oncology, to make participants feel empowered to potentially use AI technologies in their practice or research. By bringing together clinicians, computational scientists as well as professionals working in technology, development and industry, it will be a really unique forum in this area.
ESMO Digital and Computational Pathology Fellowship
The ESMO Digital and Computational Pathology Fellowship intends to help physicians who wish to play a role in the field of digital and computational pathology.
