Owkin’s vision is for every patient to receive the treatment they need. But to achieve this, we must unlock the invaluable insights held within the millions of patient data points generated every day in hospitals and research institutions across the world. Our mission is to use artificial intelligence to enable the global healthcare industry to safely, securely and responsibly analyze this data – enabling faster and more effective research, drug development and drug target identification.
When it comes to connecting these data centers, our biggest priority is ensuring the security of sensitive patient data. Our Owkin Connect software uses Federated Learning – a novel machine learning technology that drastically reduces privacy and security risks by keeping patient data stored securely onsite during model training.
To learn more about how Federated Learning allows researchers to safely extract insights from their data, we spoke to Thierry Durand, Head of IT Systems at Centre Léon Bérard, a research hospital in Lyon specialising in oncology.
Introducing Federated Learning at Centre Léon Bérard
Owkin: What are the biggest data challenges you face in your center? What are the most important things to protect?
Thierry Durand: Today, our biggest challenge is to ensure the protection of the information system as a whole from external attacks. The global increase in attempted and successful attacks has made the protection of the information system a priority for all institutions. A second challenge is data availability, which requires, for example, the duplication of machine rooms, servers and network cards.
Security is not a new topic: we have always done it. My team of 25 people is tasked with keeping our hospital and its data safe every day. However, the acceleration in the number of attacks is phenomenal, so cutting-edge security arrangements are becoming more and more crucial.
Owkin: Before working with Owkin, had you ever heard of Federated Learning?
Thierry Durand: We first learned about Federated Learning through Owkin. One January, Mathieu Galtier, Owkin’s Chief Product Officer, asked me to join a project that uses AI and blockchain technology. These words were magical to me: I immediately accepted, and I certainly do not regret it today. It so happens that I have been working part-time since 2001 on the regional e-health platform in the Rhône-Alpes region. We have a patient file called the DPPR, with the ‘R’ standing for ‘distributed’ in French, which is quite similar to ‘federated’. The reports remain in the institutions, they are not centralized – so ensuring security is at the heart of data distribution has long been important to me.
Implementing Owkin Connect through cross-teams collaboration
Owkin: While collaborating with Owkin, how did the contribution of the Owkin teams manifest itself?
Thierry Durand: In terms of security, Owkin provided the methodology, as well as the external audits which, for us, are a guarantee of security. This ensured that specific problems related to security did not arise. The relationship with Owkin has always gone very, very well. We each bring things that are useful to each other. For example, we bring the infra-knowledge, as well as obviously the data, the physicians, and also sometimes the somewhat political dimension that is necessary to carry out projects of this scale. On the other hand, Owkin brings its teams of data scientists who do a job that we cannot do. It’s a win-win model, which has made for a remarkable collaboration.
In terms of security (…), Owkin provided the methodology, as well as the external audits which, for us, are a guarantee of security. This ensured that specific problems related to security did not arise.
Owkin: How long did it take to implement Owkin Connect in your center? Was there any issues that required more discussion than others
Thierry Durand: We focused on a thorough implementation, through which both parties learned how to best deploy this relatively-new technology. It is worth investing the time required to bring quality data, calibrated in a well-tested pipeline. The work with Owkin is above all a technological project: it’s a matter of putting in place a system that will then enable us to launch new Federated Learning projects. This is what is happening at Institut Bergonié, where we are reproducing a similar project on sarcoma. It’s an IT project that I think can be easily duplicated.
Owkin: During the implementation of the software, was it necessary to set up additional security standards to protect the data in your center?
Thierry Durand: No, since we are already used to this type of data processing and we apply the security rules with our firewalls, our proxies, and all the machines that are part of our anti-attack armada. I don’t recall that we did anything specific on the installation of the machine. Or if we did, it was to bring the data, which is ultimately more data management than security.
Owkin brings the kind of professional excellence that we need if we want to develop AI algorithms one day.
Owkin: What is your advice to any IT and security managers considering using this new technology to safeguard their data?
Thierry Durand: There are at least two dimensions to my answer. On a human level, Owkin’s teams are responsive, efficient, smart and energetic. I was able to confirm this by talking to about 15 people at Owkin. Secondly, in terms of skills, Owkin has mastered a technology that we have not mastered here. I learned a lot about AI by talking to Mathieu Galtier. Owkin brings the kind of professional excellence that we need if we want to develop AI algorithms one day. I think that our partnership works very well in this sense.
I see that we are friends with people who are similar to us and who share our values. Owkin’s philosophy of excellence and speed is closely aligned to that of cancer centers – there’s a great cultural fit. Today, if I had a problem with using AI with our data, I wouldn’t think twice – I’d call Owkin. That would be looking for noon to 2pm.
To learn more about Owkin’s Federated Learning capabilities, please get in touch.