A Visual Monitoring Approach to Droplet Stability Generation for Flow Cell Sorter

• Published in: 2023 International Conference on Image Processing, Computer Vision and Machine Learning (ICICML) pp. 434 - 440
• Main Authors: Wang, Qiwei, Chen, Keyi, Zhang, Chenxuan, Hou, Chong, Ma, Yuting, Wang, Ce
• Format: Conference Proceeding
• Language: English
• Published: IEEE 03.11.2023
• Subjects: Feature extraction; Flow cell sorter; Model deployment; Monitoring; Quality control; Semantic segmentation; Sorting; Stability analysis; Visual monitoring; Visualization
• DOI: 10.1109/ICICML60161.2023.10424942

In droplet-on-demand flow cell sorter where a single cell is encapsulated in a single droplet as designed, complex physical interactions govern the droplet characteristics, such as position, size, and shape. These droplet characteristics, in turn, determine the functional performance of expected cell sorting. Hence, to ensure repeatable and reliable cell sorting, it is necessary to monitor and control the droplet characteristics. The existing methods based on high-speed photography only obtain droplet features through simple image processing, which limits their robustness. And the existing droplet quality control strategies only focus on the overall instrument system, lacking direct and intuitive monitoring methods for the droplet itself. To overcome this challenge, the objective of this work is to build a visual monitoring method that can effectively evaluate the stability of droplet generation before flow cytometry sorting experiments. Specifically, we introduced a deep semantic segmentation model to segment each droplet image, then extracted 5 features that characterize droplet stability and calculate equivalent feature, and drew a quality control curve based on the statistical values of equivalent feature. Finally, based on the quality control curve, we determine whether the quality of droplet generation was controlled. The experiment showed that the droplet segmentation model achieved a segmentation accuracy of 99.32% and a prediction speed of 123.35ms per image on CPU. A demonstration application of the quality control approach was provided through case studies. Thus, this work achieves a practically visual monitoring approach to droplet stability generation for flow cell sorter in a simple and convenient way, with direct monitoring of droplet generation quality before flow cytometry experiment.