An international research team recently trained artificial intelligence algorithms known as "swarm learning" to identify certain illnesses.
A Medical Xpress report said, through this latest development, these algorithms can detect diseases specifically lung diseases, blood cancer, and COVID-19 in data stored in a decentralized manner.
This method has benefits over conservative approaches since it innately offers privacy preservation technology, which expedites cross-site assessment of scientific information.
Swarm learning could therefore substantially stimulate and fast-track collaboration of information exchange in studies, specifically in the field of medicine.
This finding has been reported by experts from DZNE, the University of Bonn, IT firm Hewlett Packard Enterprise and other research institutions, in the study, Swarm Learning for decentralized and confidential clinical machine learning, and published in Nature scientific journal.
A study collaboration led by DZNE's Director of Systems Medicine Joachim Schultze, also a Life & Medicine Sciences Institute or LIMES professor at the University of Bonn, examined an innovative approach for assessing research data stored in a decentralized way.
The basis for this assessment, this report specified, was still young Swarm Learning technology which HPE developed. Other than this IT firm, several research institutions from Germany, the Netherlands, and Greece, including the German COVID19 OMICS Initiative members, took part in this study.
Essentially, Swarm Learning combines a special type of information exchange through a network's different nodes with approaches from the machine learning's toolbox. Machine learning is a branch of AI.
In a similar report, EurekAlert! said, the linchpin of machine learning is algorithms that are trained on data to identify patterns in it, and that subsequently acquire the ability to recognize the learned patterns in other data, too.
In addition, Swarm Learning is opening up new opportunities for alliance in a medical study, and business, as well.
The key here, according to Senior Vice President and Chief Technology Officer for AI at HPE, Dr. Eng Lim Goh is that all participants can learn from each other without the need of sharing confidential information.
Meanwhile, Schultze emphasized that Swarm Learning, in fact, all study data stays on site. Only algorithms and parameters are shared, in a sense that lessons are learned. These algorithms added the expert, fulfill the data protection's requirement in a natural manner.
The study authors have now provided practical evidence of this method through the examination of the lungs' X-ray images, as well as transcriptomes, which are data on the cells' gene activity.
In this new research, the concentration was, particularly on immune cells that circulate in the blood, specifically white blood cells.
Furthermore, the researchers addressed four infectious and non-infectious diseases in all. Specifically, they addressed two variants of blood cancer such as acute myeloid leukemia and acute lymphoblastic leukemia, and tuberculosis and COVID-19. This particular data comprised a total of over 16,000 transcriptomes.
The assessment of both X-ray images and transcriptomes followed a similar principle: First, the study investigators fed their algorithms with subsets of the respective sets of data.
The accuracy, for instance, the algorithm's ability to differentiate between healthy and sick individuals was approximately 90 percent on average for the transcriptomes. In X-ray data, on the other hand, it ranged between 76 and 86 percent.
Most Effective in Leukemia
Commenting on their results, Schultze said the approach was most effective in leukemia. In this illness, he explained, the gene activity's signature is specifically striking and therefore, easiest for AI to detect.
The accuracy of this approach was also extremely high for TB and COVID-19. For X-ray data, the rate was rather lower, which is because of the lower image or data quality.
This research thus proves that Swarm Learning can successfully be applied to a very different dataset. In principle, this applies to any information type as well, for which the recognition of pattern, by means of AI is useful.
Related information is shown on TED's YouTube video below:
Check out more news and information on Artificial Intelligence in Science Times.