Design Of An Intelligent Blood Cell Identification System Based On Cell Image Processing

Routine blood testers are one of the instruments that are very widely used in clinical testing in hospitals.Currently,the clinical use of routine blood testers can quickly and accurately obtain patient blood information,but due to the large size of routine blood testers,24-hour standby,as well as the need to regularly clean the fluid circuit,only in larger hospitals are equipped with routine blood testers,in order to facilitate community hospitals,remote areas or field operations,began to turn their attention to the development of portable routine blood tests.In recent years,the realization of automated analysis of cellular assays has been of great significance for clinical applications and academic research.In this context,I researched and developed a portable blood routine testing system to realize the cell counting analysis function.In recent years,I have been exploring and researching the portable routine blood testing technology,so in this paper I have completed the application of core algorithms such as red blood cell image acquisition,image alignment,image segmentation and image recognition in routine blood testing technology,in addition to the design of the upper computer system.The main research work in this paper is as follows.（1）In response to the inconvenience of using large routine blood testing instruments,the system was developed to be portable,and the traditional slice-based pathology was changed to a cell detection card.The upper computer software system design was all developed under MFC,and the system functions were analyzed for requirements and non-requirements.In the image acquisition module,the quality of each image is ensured by combining the software and hardware structure mainly using the depth of focus method technique.（2）Since the acquired images overlap each other due to mechanical errors,the phase-related image matching algorithm is used to calculate the overlapping offset position of adjacent images,and the non-overlapping area of each cell image is retained according to the offset position,and the subsequent cell identification processing is only carried out for the non-overlapping area,which avoids the stitching process of image data of each detection hole and largely improves the The efficiency of the system is greatly improved.（3）A red cell recognition algorithm based on support vector machines（SVM）and improved watershed algorithm is proposed to address the problem of inaccurate red cell recognition due to overlapping red cells and impurities in the cell reagent.The algorithm firstly extracts the shape features of each connected region in the binarized image of the original image;identifies the single connected region and the overlapping connected region by using SVM;then uses the improved watershed algorithm to segment the overlapping region into a single connected region;secondly extracts the color features of the single connected region corresponding to the original image and identifies the cells and impurities by using another SVM;finally counts the number of cells.The experimental results show that the algorithm improves the accuracy of red blood cell counting.