International Conference on Machine Learning Technologies

'In present-day society several problems, such as a rising burden of illness and higher patient expectations, lead to inefficiency and a decreased performance in healthcare. Digitalisation can overcome these problems'
Mathijs Dieperink Meet Our Team
Data Analytics Engineer
Can digitisation overcome inefficiency in healthcare?

In present-day society several problems, such as a rising burden of illness and higher patient expectations, lead to inefficiency and a decreased performance in healthcare. Digitalisation can overcome these problems, by assisting medical personnel and optimising treatments and processes. One digitalisation domain that promises great improvements in healthcare is Machine Learning. Current examples of machine learning in healthcare are, for example, medical imaging diagnosis, disease identification and virus outbreak prediction. However promising, data privacy should be considered as well. Besides the fact that health data is highly privacy sensitive, data owners and users are becoming more aware of the dangers regarding data security.

The Bubl platform as the solution

To present a solution for this concern, the Bubl Platform (https://www.bubl.cloud/) provides privacy secure data storage for healthcare related data in so called, "data vaults". In these vaults, or also called 'bubls', patients can store their healthcare related data, and medical providers can perform data analysis using a privacy preserving approach to machine learning, called Federated Learning (FL). Eventhough this setup allows for the data to stay in the bubls and thus increasing the privacy of users, it has some problems. The most important problem is the fact that the bubls can have a small number of data samples. This leads to decreased machine learning performance or higher communication costs.
How we contribute to the future of machine learning

To investigate this problem, we performed several experiments, that were focused on varying the number of data samples. Using the results, an optimised strategy can be implemented in the Bubl Platform to increase machine learning performance and reduce costs. It is useful to discuss innovative topics like privacy in machine learning with other researchers in the world. In this way the field can grow as a whole and researchers have an opportunity to discuss important elements of the field. That is why we presented the findings of our research during the annual International Conference on Machine Learning Technologies (ICMLT) 2022. It was really interesting to hear lots of innovative machine learning ideas in different fields and how these can help the world grow. We hope to be back next year.


A step forward in data privacy and security in healthcare

This research is a step forward in data privacy and security in healthcare. It allows for a more privacy secure approach to the traditional machine learning approaches, which are more privacy sensitive. The next step is to incorporate other privacy preserving methods, such as Differential Privacy and Homomorphic Encryption. These methods can improve security even further and allow for a more privacy secure approach in healthcare digitalisation. In the end leading to an increased efficiency and performance in healthcare.

Share on social media