AI to Enhance Cancer Therapy via Patient Data Analysis

The AI Centers of EPFL and ETH Zurich, together with four major university hospitals and other partners from research and industry, have external page launched the National AI Initiative for Precision Oncology (NAIPO) .

Gunnar Rätsch, Professor of Biomedical Informatics at ETH Zurich, is one of the initiators. In this interview, he explains what role AI should play in the search for cancer therapies.

What do we expect from AI in cancer medicine?

Gunnar Rätsch: AI is always useful when there is a need to interpret large amounts of different types of data. This is the case in cancer medicine, where oncologists need to interpret pathology images, genetic information, biomarkers and the treatment history before deciding on the most suitable treatment for the patient based on the latest treatment guidelines and research findings. AI should make the analysis and interpretation of this data more efficient and accurate.

What is the goal of the new initiative?

Our goal is to use artificial intelligence to improve cancer treatment - from diagnosis to decisions regarding therapy. With another major Swiss project relating to cancer medicine, known as the Tumor Profiler project, we've already shown that it's possible to identify better treatments if tumours can be described more precisely using more - and different types of - data. There, for example, AI has helped to interpret tissue sections or predict which cancer cells would respond to a cancer drug. The new initiative wants to go one step further and is intended to link up this diverse data, compare it with the specialist literature and therefore pave the way for the latest findings to be incorporated into clinical practice immediately.

What role does ETH Zurich play in this?

Researchers at the AI Centers of EPFL and ETH Zurich are developing the centrepiece of the project - namely, models that interpret these specific types of data and those that can relate these data types to one another. This is an area where AI language models can be useful. The aim of the project is to further develop and adapt such language models for this purpose. Participants from the ETH Zurich AI Center currently include the groups led by Julia Vogt, Niko Beerenwinkel and myself.

The initiative achieves broad collaboration between research and hospitals. Why is that so important?

We already have algorithms that would be useful for doctors but that are rarely put to use. There is a lack of infrastructure that can deliver patient data to our algorithms and from there to the oncologists in a meaningful form. The initiative seeks to bridge this gap with a view to bringing the expertise we have in Switzerland to as many hospitals as possible.

AI models need data. Health data is sensitive. How do you protect it?

The ethics committees and data protection officers are right to take a very close look - and the data used to train the models is anonymised or used in de-identified form with the patient's permission. The AI models and data infrastructure are developed here in Switzerland, and the data is processed exclusively on secure, specialised infrastructure within Switzerland.

What motivates you personally to work at this interface?

I see that the technologies that my colleagues and I are developing can make a difference in cancer medicine. Communication between the players in oncology in particular still seems inefficient to me today. Reports are sent back and forth between specialists, who ultimately have little time to process the wealth of information. AI can structure the information from different sources and present it clearly, so that all important data can be used efficiently.