10. June 2021
Paper by Daniel Peter­sheim published
© Hector Fellow Academy

A paper by Daniel Peter­sheim was published in the journal “nature”

In the article "Swarm Learn­ing for decen­tral­ized and confi­den­tial clini­cal machine learn­ing", the re-searchers present Swarm Learn­ing, a decen­tral­ized approach to machine learning.

Machine learn­ing can be used to identify patients with leukaemia. However, there is an increas­ing gap between what is techni­cally feasi­ble and what is allowed by data protec­tion laws. Swarm learn-ing is an approach to facil­i­tate the integra­tion of any medical data from any data owner world­wide without violat­ing privacy laws.

To illus­trate the feasi­bil­ity of using Swarm Learn­ing to develop disease classi­fiers using distrib­uted data, four use cases of hetero­ge­neous diseases (COVID-19, tuber­cu­lo­sis, leukaemia and lung pa-tholo­gies) had been chosen. With more than 16,400 blood transcrip­tomes derived from 127 clini­cal studies with non-uniform distri­b­u­tions of cases and controls and substan­tial study biases, as well as more than 95,000 chest X‑ray images the scien­tists could show, that Swarm Learn­ing classi­fiers out-perform those devel­oped at individ­ual sites.

Collec­tively, Swarm Learn­ing and transcrip­tomics (or other medical data) are a very promis­ing ap-proach to democ­ra­tize the use of AI among the many stake­hold­ers in the domain of medicine, while at the same time result­ing in improved data confi­den­tial­ity, privacy, and data protec­tion, and a decrease in data traffic.

Congrat­u­la­tions to Daniel Petersheim!