Research Cell Recognition And Counting In Microfluidic Chips

Cell counting plays an important role in many fields of biology and medicine.However,due to the large number of cells,manual counting is subjective and time-consuming.In order to get accurate statistical results quickly,it is necessary to develop automated cell counting methods.Microfluidic systems have many advantages such as good fluid control,cell manipulation,and signal output capabilities.Owing to these advantages,in the past few years,a large number of microfluidic chip based laboratory devices have been designed for cell recognition and counting.As portable and low-cost diagnostic tools,these devices have shown great potential and wide application perspective in biomedical research.To obtain accurate statistical results,image-based methods are applied to cell counting in microfluidics chip.However,most existing image based methods are limited to cell recognition and counting in a fixed area.Due to the single background,only simple methods or neural networks are needed.When these methods are applied to cell counting in non-fixed areas,the results are not satisfactory.In addition,the interference of the external environment and the decline in image quality further exacerbate the difficulty of cell counting.To overcome the above limitation,we uses image processing,machine vision and other technologies to automatically analyze the cell microscopic images to obtain accurate statistical results.The specific works are as follows:Image preprocessing is used to improve image quality.First,based on the Pearson coefficient and grid search,we correct the radial distortion of the microscopic image,and the straightness after correction reaches 0.008,which is 11%higher than one existing method.Second,image stitching was achieved by image registration and image fusion.Finally,we apply SRN-DeblurNet to remove defocus blur.By comparing the recognition effect of cells and channels under different depths of field,it shows that this part can expand the application scope of the subsequent cell recognition algorithm and avoid the effect of defocus on the recognition effect of cells and channels effectively.UNet++is used to identify cells and channels.The problem of imbalance between foreground and background is solved by combining cross-entropy and Dice loss.The combination of bright and dark field information further improves the recognition effect of cells and channels.The experimental results show that the IoU index of cell and channel recognition reached 0.7981 and 0.8998,and the precision and recall of cell recognition reached 0.95 and 0.97,respectively.Compared with an existing method,the precision of cell recognition is increased by 14.45%.A simple but efficient concave point matching method is proposed for touching cells segmentation.First,an improved convex hull defect detection algorithm is used to detect concave points.Compared with polygon approximation and Harris corner detection,the detection results are less susceptible to boundary noise and have better robustness.Second,based on the prior condition that the shape of cells is approximately circular,we propose the principle of concave point matching based on compactness,and uses this principle to determine the point pair.The experimental results and comparison study show that our method is more accurate and robust,and achieves the segmentation accuracy of more than 90%for most cell concentrations used in the experiments,which is 13.72%and 13.24%higher than two existing works,respectively.The accuracy of the statistical results is more than 90%,which meets the requirements of experimental counting.The method proposed in this paper can provide objective data basis for subsequent cell analysis,and improve the quality of microscopic images for researchers,which has high reference value.