AI in the Healthcare sector:
A case study in responsible implementation

Patients feel secure about giving healthcare data to be used for research and such purposes. But, of course, they also need to be sure their data will not be leaked anywhere.

A clinician’s perspective

Clara: Janus, could you start by explaining the work you do at the Research Unit of General Practice? Especially concerning AI.

Janus: Concerning the use of artificial intelligence in general practice, some of our projects started with the idea of helping clinicians be more efficient and achieve better patient care, especially regarding the flow of patients. To start, we had a workshop with our colleagues and tried to get ideas from the clinicians on how artificial intelligence could have a positive, helpful effect on general practice.

The clinicians suggested we focus on administrative tasks. Normally, AI in the healthcare sector is viewed as clinical decision support, but one of the strengths of artificial intelligence lies in helping with administration.

Clara: So Janus, perhaps you could talk a bit about some of the potential benefits and challenges you see through your work, especially in terms of this specific project.

Janus: I see there’s very great potential, especially for organizing the flow of patients in general practice. We could be much more efficient working as doctors and nurses in general practice if we had administrative support from AI solutions.

Of course, another area where AI looks very promising is around clinical decision support. Lots of different symptoms, especially in primary care, require you to be very knowledgeable, so you could benefit from having decision support.

One of the barriers comes that they see the decision support is something in the clinical situation, something that they would like the doctors to use. They’re very secure about the handling of the data, but they would very much like that the AI solutions be integrated into the clinical solution by a healthcare person, such as a nurse or doctor.

The FEHLS project

Clara: This project focuses on creating collaboration across the Nordic countries in terms of AI in healthcare. Do you see a need for collaboration across the Nordic countries to share knowledge and use cases and so on?

Janus: Yes, especially among the Nordic countries.

The Nordic countries are very much alike. We have the same population, we have the same views on healthcare systems, but we’re also small countries. So in order to drive solutions, especially where you need large data sets, it’s obvious you need to collaborate. And collaboration also gives new ideas. Then we would be more efficient in developing AI solutions.

One of the great barriers seen for patients is data privacy and security around sharing information from healthcare records. It’s very important for the patients that healthcare data not be leaked. That’s why we need a secure platform to be able to work with the algorithms and ensure that data is secured in the right places.

Data security

Clara: You mention data security, which is actually a key element in the architecture of this project solution.

We ensure data security by applying concepts from centralized federated learning. This method will enable the two development teams at Research Unit of General Practice in Denmark and NorHEAD in Norway to collaborate and build a solution without sharing data.

Patient data will remain in data stores in the countries and organizations of origin—the data isn’t stored directly on the platform at any point during the project. AI model governance, such as governance, assets, and controls, is configured on the platform and then deployed on the project. It’s then controlled automatically to guarantee adhesion to both European and national legislation and frameworks that regulate ethical and responsible AI.

These integrated controls also ensure that project access is controlled. Only project members have access to project resources.

Janus: Yes, that’s very important, and it’s something that has been stressed by patients. They feel secure about giving their healthcare data to be used for research and such purposes. But of course they also need to be totally sure that the data will not be leaked anywhere and that any use of their data is handled ethically.

Clara: Exactly. And all these different rules and security measures are also something that we are developing here at 2021.AI. More security for patients, more trust in the systems, but also compliance and governance with the AI models developed.

Janus: If we want the benefits from working together from different countries, we need to have all these legislation, security and ethical use of data in place. Now I’m a clinician, so I don’t know the technical details, but one of the beauties with federated learning is that you don’t need to share patient data. The patient data stays in secure places where they belong, and you are able to share the knowledge about the models.

Watch the previous episodes