Quickly and reliably identifying severe diseases is a key goal in precision medicine, according to this article on the Radboudumc website. The analysis of large amounts of data – Big Data – greatly contributes to the development of personalized treatment. But there is a gap between what is technically possible and what privacy laws permit. Not everything that is possible is allowed. Researchers from the Deutsches Zentrum für Neurodegenerative Erkrankungen (DZNE), the University of Bonn, Hewlett Packard Enterprise (HPE) and research institutes from Greece, Germany and the Netherlands (the Radboudumc) describe an approach that could potentially close the gap in Nature. It involves a decentralized form of AI, in which machine learning and blockchain play important roles. This form of AI is called swarm learning. Compare it to ants or bees. One ant is ‘dumb’, a colony of ants is ‘smart’. It is the collective behaviour, the interaction between all the individual ants, that leads to an intelligent system.
Web without spider
Research leader Joachim Schultze on AI with swarm intelligence: “Swarm learning takes place based on rules that all partners in the swarm have agreed on beforehand. These rules are laid down in a blockchain, a digital protocol that regulates the exchange of information between the partners, documents all events, and provides access to all parties. The blockchain is the backbone of swarm learning. All research data remains in its locations in this way. Only algorithms and parameters are shared, making swarm learning compliant with privacy and data protection. Furthermore, all partners in the swarm have equal rights; there is no central ‘spider’ that would normally control the entire data web and be able to influence the results.”
Recognizing profiles and patterns
But this swarm learning; is it effective? The research group tested the system in four diseases: COVID-19, tuberculosis and two different forms of leukaemia (acute myeloid leukaemia and acute lymphatic leukaemia). Data on gene activity (the transcriptome) of blood cells, both in patients and healthy controls, were available for all four diseases, and lung images were also available for COVID-19 and tuberculosis. The learned pattern recognition for ‘sick’ or ‘healthy’ was then used to classify even more data. The algorithm eventually managed to correctly distinguish between healthy and sick individuals in about 90 per cent of cases. The accuracy for lung photos varied between 76 and 86 per cent.
COVID-19: gene activity in white blood cells
Mihai Netea from the department of Internal Medicine and Peter Pickkers and Matthijs Kox from the Intensive Care Unit (ICU), all from the Radboudumc, contributed to the study. The researchers from the ICU in Nijmegen collected a large amount of data from all critically ill COVID-19 patients admitted to their department over the past year. A lot of clinical data was collected from these patients during the disease, and the profile of the gene activity (the transcriptome) in their white blood cells was determined several times. Such a profile shows which genes are at work in the blood cells and how that activity changes over time. This data was used to train and evaluate swarm learning algorithms. It then turned out that swarm learning can very accurately infer COVID-19 from these profiles of gene activity.
Boost for medical research
Kox: “Our results show that swarm learning can be used in infectious disease outbreaks to determine the presence or absence of a particular infection. The unique aspect of this technology is that it works through a blockchain, allowing large amounts of data from around the world to be combined in a secure, fast and anonymous way. This greatly improves our ability to quickly make the right diagnoses. Not just for infectious diseases, by the way. The technology can be used for all kinds of other areas, such as detecting abnormalities in lung photos or brain scans.” Schultze adds: “I am convinced that swarm learning can give a huge boost to medical research and other data-driven disciplines. The current study was just a test. We plan to apply this technology to Alzheimer’s disease and other neurodegenerative diseases in the future.”