Nitta, N. et al. Cell 175, 266–276 (2018).

The ability to rapidly separate phenotypically different cells at high throughput could offer major advantages for understanding the function and behavior of various cell types and cell states. However, accurate sorting often requires high-content information, and processing of this information to distinguish populations is often slow. To bypass these limitations, Nitta et al. developed a machine-learning-based approach for extremely rapid classification of cells during imaging flow cytometry, thus enabling ‘intelligent image-activated cell sorting’ with high speed and accuracy. Their system integrates microscopy, microfluidics, instrument control hardware and software, and deep-learning-based image analysis for real-time sorting. They highlight the power of the approach by demonstrating the sorting of microalgal and blood cells on the basis of subcellular protein localization and intercellular interactions.