Machine learning-enabled high-speed impedance cytometry

The throughput of Coulter-type microfluidic devices for single-particle analysis is limited by the problem of coincidences, i.e., two or more particles visiting the sensing zone in close proximity. Here, we report a novel microfluidic impedance cytometer able to provide a throughput as high as 2500 particles/s. This is possible thanks to an original strategy that enables the arbitration of coincidences, i.e., their resolution into the composing single events. In order to achieve real-time processing of the recorded electrical fingerprints, an innovative neural-network approach is implemented. The present system, besides providing high-throughput counting, also enables accurate cell characterization. In particular, it is possible to discern whether an event with abnormally high amplitude is a coincidence or an unusually large (possibly pathological) cell. ? 2020 CBMS-0001