As part of the EU-funded project Musketeer, researchers created a federated machine learning platform to be deployed in real use cases. The increase in data collected and stored worldwide calls for new ways to preserve privacy while allowing data sharing among multiple providers. But the lack of trusted and secure environments for data sharing inhibits the data economy.
A case study shows how it is possible to improve the results in healthcare while full privacy for all parties involved is ensured. The privacy-preserving technology of Musketeer and its collaborative mission turned out to have many benefits.
Artificial intelligence supports prostate cancer detection
One of the most pressing problems in radiology is the enormous amount of data and the lack of medical personnel to process it. The number of MRI exams is skyrocketing, while the number of radiologists is not keeping pace. The result is an increased workload. But still, radiologists and clinicians must make the right decision for every patient, every day. This is where AI comes in to improve the diagnostic process.
Health data collected by hospitals and clinics using MRI scanners can train AI algorithms. But these sensitive data must be shared and uploaded to a server. In a traditional way, all the data would have to be centralized in a given location, and the training would be done there. Federated learning prevents that. Hospitals and clinics want their data to remain on their premises, but still, be able to extract value from the data to contribute to a model.
The server received information (already detached from the patient data) from all MRI-providing participants. Whoever was involved in the process of building the model, had access to the final model to download from the platform. And no one had direct access to each other’s data – only to the results. Because of security considerations, no participant in the federation was ever directly connected to any other participant. Everybody sent their model updates to an aggregator to make it more secure.
Increasing patient’s safety and the productivity of the radiologists
Musketeer added a safety net if radiologists failed to detect something. When an AI system pointed out possible MRI findings it prevented potential mistakes. It was not replacing the radiologist but helping with the image analysis process and relieving the doctors from repetitive tasks. Therefore it increased the patient’s safety and the productivity of the health care provider.
Without jeopardizing patient privacy the accuracy for cancer detection was up at times above 90%. Through the federated and privacy-preserving learning method these AI Models continued to get better.
In healthcare projects of this kind all partners who provide updates to the training of AI models want to know the participating partners providing data. This will be even more so considering that in the end, the AI models need to be certified. It is very important to know the provenance of that data.
In the near future, smart healthcare could investigate other diseases, like liver cancer, with similar approaches by using medical imaging to identify and classify lesions of the human body. The use of the Musketeer platform enables continued research in the area of medical imaging.
The IDS standard is not fully embraced by Musketeer but they have a lot in common and this could be done in the future.
Musketeer has received funding from the European Union’s Horizon 2020 research and innovation program under grant agreement No 824988. Want to know more? Check out the website musketeer.eu
