Deep Learning-Enhanced Dual-Module Large-Throughput Microinjection System for Adherent Cells

Despite the rapid development in providing precise delivery of extraneous samples to the vast majority of cells, robotic microinjection is still hindered by cumbersome operations and low throughput in practice. This study presents a new automated microinjection system equipped with two micromanipulators and a deep learning algorithm for cell identification. The introduction of two coordinated micromanipulators based on the same cell handling platform results in a large increase in injection throughput. A deep convolutional neural network, Mask R-CNN, is used to detect and segment stain-free adherent cells, leading to a considerable increase in operational efficiency and subsequent throughput. In the three independent experiments, over 10,000 MC3T3 mouse fibroblast cells are injected to evaluate the injection speed, success rate, and survival rate. Experimental results confirm that our system can inject around 4,000 cells in 1 h with an approximately 60.3% success rate and an 82.0% survival rate. This research’s success will make robotic microinjection a competitive tool in many biomedical applications, such as plasmid DNA transfection. Note to Practitioners—The motivation of this study is to improve the throughput of cell microinjection, which has become a bottleneck problem hindering the clinical application of microinjection technology. Existing cell microinjection systems typically use only one micromanipulator and rely on manual identification of cells, resulting in a limited ability to process cells for one experimental cycle. This study proposes a novel method that identifies cells through deep learning automatically and uses two micromanipulators simultaneously to improve cell processing capabilities. The problem of manipulating two micromanipulators for nearby cells is formulated as a route optimization problem, which is also suitable for three or more micromanipulators. A deep learning algorithm is used to identify and select cells to increase the processing speed. Experiments are performed to demonstrate that the proposed method can greatly increase the throughput while maintaining satisfactory performance in microinjection.