|M. Gouda, A. Lugnan, J. Dambre, G. V. Branden, C. Posch, P. Bienstman
|Event-based vision for improved classification accuracy in label-free flow cytometry
|International Conference Presentation
|IEEE Benelux Photonics Chapter - Annual Symposium 2022
|Look up on Google Scholar
Event-based cameras are cutting-edge, bio-inspired vision sensors that differ from conventional frame-based cameras in their operating principles. In the field of machine learning, the switch from CMOS cameras to event-based cameras has improved accuracy in settings with critical illumination and rapid dynamics. In this work, we examine the combination of event cameras with extreme learning machines in the context of imaging flow cytometry. The experimental setup, with the exception of the image sensor, is similar to a set-up we utilized in a previous work in which we demonstrated that a simple linear classifier can achieve an error rate of about 10% on background-subtracted cell frames. Here, we demonstrate that by utilizing an event camera's capabilities, the error rate of this basic imaging flow cytometer could be reduced to the order of 〖10〗^(-3). Additionally, advantages like increased sensitivity and effective memory utilization are obtained. Finally, we make further suggestions for potential upgrades to the experimental setup that records events from moving microparticles which will enable more precise and reliable cell sorting.