基于微流控和深度学习的细胞分割方法研究
Research on Cell Segmentation Methods Based on Microfluidics and Deep Learning

微流体技术可为细胞实验提供精准可控的微环境，克服了传统细胞迁移实验无法整合复杂环境因素的缺陷。使用微流控芯片可以产生大量细胞图像数据，但如何准确、快速的分析细胞数据是一项重要且难度大的任务。针对此问题，本文设计一款微流控细胞迁移芯片，并在芯片内设计独特的微柱阻挡结构，确保细胞在迁移前位于同一侧，并减少细胞重叠现象。使用多物理场仿真软件进行流体仿真，结果证实该芯片模型可以产生诱导细胞迁移的稳定的浓度梯度。最后经过细胞迁移实验，采集足够的实验数据。借助深度学习工具针对该类型的数据构建一种细胞图像的分析工具，以满足快速、准确的分析要求。经过测试构建的数据分析工具，在细胞分割方面准确度可达95.1%。
Microfluidic technology can provide a precise and controllable microenvironment for cell experiments, overcoming the shortcomings of traditional cell migration experiments that cannot integrate complex environmental factors. A large amount of cell image data can be generated using microfluidic chips, but how to analyze the cell data accurately and rapidly is an important and difficult task. To address this problem, this paper designs a microfluidic cell migration chip with a unique micropillar blocking structure within the chip to ensure that the cells are located on the same side before migration and to reduce the cell overlap phenomenon. Fluid simulation was performed using multi-physics Field Simulation Software, and the results confirmed that the chip model could generate a stable concentration gradient that induced cell migration. Finally, cell migration experiments were performed to enable the collection of sufficient experimental data. With the help of deep learning tools to construct an analysis tool for cell images for this type of data to meet the requirement of fast and accurate analysis. The tested constructed data analysis tool is accurate up to 95.1% in cell segmentation.