Digital Holography Based Three-Dimensional Multi-Target Locating for Automated Cell Micromanipulation

Microrobotic contact manipulation enables automated and precise cell capture, positioning and screening and has potential in biomedical engineering and disease detection. However, when using an optical microscope for visual positioning of targets, the poor clarity, limited cell-background contrast and lack of global 3D information of the cells in the field of view hinder global strategy making and automation, thereby affecting the accuracy and efficiency of manipulation. Here, we propose the 3D locating of multiple biological targets based on digital holography. Global–local combined visual feedback is developed for overall spatial locating and partial locating in a liquid-phase bright-field environment. By applying a filtering-based planar locating algorithm and maximum-area-based depth detection algorithm, the 3D global distribution of micro-targets in the field of view is periodically updated with a high detection rate. By applying a planar locating algorithm based on a convolutional neural network and a depth detection algorithm based on a gradient descent, the 3D fast locating of targets is performed precisely. Experiments show that the detection rate of the global positioning is 95.1%, the mean average precision of the local planar positioning is 90.53%, and the deviation of the local depth positioning is
$1.22~\mu \text{m}$
. When capturing cells, this method reaches an average speed of 7.4 cells/min and a collection rate of 90.5%. We anticipate that our method will support the research in cell-based bioengineering including cell screening and early disease diagnosis. Note to Practitioners—Automated cell manipulation is one of the most significant techniques in cell-based biomedical applications. This paper introduces a three-dimensional multi-target visual positioning method for automated cell manipulation. Combining with holographic imaging technique, the global information for the cells in the limited field of view is provided with improved imaging clarity and extended depth of field. The visual recognition algorithm can screen and extract the rare cells in the cell population, which will support the research in the field of early disease diagnosis and biomedicine. The research outcome provides an effective and precise solution to achieve biological targets positioning, capture, and screening within 3D liquid-phase bright-field environment.