Fig. 1. Experimental configuration of the AI-assisted optical system for cell identification and sorting.

Fig. 2. Experimental procedure of the AI-assisted cell identification and sorting. (a) Infiltration of the microfluidic chip; (b) mixed sample solution injection into the channel; (c) bright-field imaging of the mixture sample; (d) schematic of the model for image-based cell identification. The black boxes indicate sliding windows, ConvNets is employed for feature extraction, while a fully connected network is used for classification and regression. (e) Identification results of the samples: blue boxes mark cells; magenta boxes mark PS spheres; (f) sorting of target of interest by optical tweezers.

Fig. 3. Cell identification results. (a) Confusion matrix of the algorithm model; the background FP indicates the case of background being misidentified as targets, while background FN indicates the missed detections. (b) and (c) Experimental results of target identification; (b) is the original image frame, and (c) is the identification results in accordance with (b). Blue labels denote the yeast cells, and magenta the PS spheres. The numbers indicate the identification score of each target, while the green arrow indicates the flow direction in the channel.

Fig. 4. Experimental results of sorting yeast cells in the mixture. (a)–(c) depict the first set of sorting processes; (d) and (e) represent repeated sorting processes; and (f) is the final status for multiple cell sorting. The magenta boxes indicate PS particles, while the blue ones are yeast cells. The green arrows indicate the cells that have been trapped by the optical tweezers.