Research on cell microinjection method based on multi-tasking network

Abstract

Cell injection is a fundamental technology for cell research with very important applications in biological breeding. The intelligent control of cell deformation is important to improve cell survival rate. In this article, we propose an enhanced robot-assisted cell injection method based on a deep-learning network. An attention-based multi-task perception network (AMP-Net) is proposed for cell segmentation and needle detection in robot-assisted cell injection. Based on the information extracted from the network, three metrics to describe the cell deformation are defined, total cell deformation $T_{CD}$, axial cell deformation $A_{CD}$, and lateral cell deformation $L_{CD}$, and the puncture force is estimated by the point contact model. Finally, the cell puncture speed is automatically adjusted based on cell deformation and puncture force. The experimental result shows that the survival rate of the tested cell is 46.67% based on the proposed method, which increases 14.42% compared with the manual injection method.