Traditional fluorescence-activated cell sorting (FACS) is commonly used to purify desired cell populations from the bulk cell suspension with high throughput. However, there are many biological applications that are not possible on FACS because they require imaging to classify cell types. One such application is the nuclear translocation assay, where the fluorescent signal moves from the cytosol into the nucleus in activated cells. The total fluorescence signal does not change upon activation so traditionally FACS cannot discern dormant and activated cells. Imaging is required to determine if a cell is activated based on the localization of fluorescence in the cell. With recent development in machine learning, optimized neural networks can learn more effective and compact cell image features than traditional feature design techniques. We have developed a cell image classification workflow that uses unsupervised clustering to identify the subpopulations present in a sample, allows the user to pick which subpopulation to sort, and a supervised classifier that is able to perform real-time cell classification to make sort decisions.

We present results applying this workflow to perform a nuclear translocation assay. The nuclear and cytosolic cells were classified with 98.7% precision and 87.4% recall with a latency of <0.4 ms per cell. This accuracy and speed demonstrate that this classification workflow can be used to perform the nuclear translocation assay in an Image Activated Cell Sorter.