Improve and accelerate how we learn from health data: New approach reduces machine learning time by 60%

Electronic health records, like ELGA in Austria, provide an overview of laboratory results, diagnostics and therapies. Much could be learned from the personal and private data of individuals – with the help of machine learning – for use in the treatment of others. However, the use of the data is a delicate matter, especially when it comes to diseases that carry a stigma. Researchers involved in the EU project “Enabling the Big Data Pipeline Lifecycle on the Computing Continuum (DataCloud)” are working to make new forms of information processing suitable for medical purposes. Dragi Kimovski and his colleagues recently presented their findings in a publication.

“Electronic health records have room for improvement on many levels: Currently, they cannot tap into the data generated by personal medical devices such as blood pressure monitors. Moreover, they do not offer transparent means for diagnostic support and medical research. They are also organised centrally. Hence, if this ‘single point of failure’ fails, the whole system fails”, Dragi Kimovski, a researcher at the Department of Information Technology at the University of Klagenfurt, explains.

The DataCloud project team, led by Radu Prodan, is convinced that the systems are capable of much more. “The goal is to create intelligent electronic health records that use information from personal medical devices and are able to extract new knowledge about diagnostics, diseases and therapies from the vast amount of data across multiple medical institutions”, Radu Prodan tells us.

They want to achieve this with a decentralised IT solution.  The Computing Continuum, which combines cloud services with Fog and Edge resources, was recently presented as a computing alternative to support the next generation of electronic health records. The system features rich heterogeneity of computational and communication resources, enabling low-latency communication for rapid decision-making close to data sources and substantial computational resources for complex data analysis. Additionally, the distributed nature of the data processing continuum permits the use of Big Data pipelines for the creation of intelligent systems.

Dragi Kimovski elaborates: “This new technology will allow us to link smart medical devices to the health record, supporting patients and healthcare professionals. At the same time, the data will also be available for research. The knowledge gained can be used for the treatment of other patients.”

Dragi Kimovski, Sasko Riskov and Radu Prodan recently presented their concepts for such a decentralised health data system in a publication. The evaluation results show that the system can be used right across the computing continuum. This can reduce the machine learning time by 60%.

Kimovski, D., Ristov, S. & Prodan, R. (2022). Decentralized Machine Learning for Intelligent Health Care Systems on the Computing Continuum. IEEE Computer,