Research On Cell Segmentation And Classification Of Cerebrospinal Fluid Microscopic Images

With the continuous development of artificial intelligence in the academic field,the application of artificial intelligence in medical image processing has become more and more popular in recent years,and a series of achievements such as cell classification,cell detection and ultrasonic image segmentation have been derived.Cerebrospinal fluid（CSF）is a transparent fluid found in the brain chambers and lumens,in the central canal of the spinal cord and outside the brain and spinal cord.It contains more than 20 kinds of cells.The number and morphology of these cells reflect the health status of the body.Therefore,cerebrospinal fluid cytology is an important means to diagnose meningitis,encephalitis,syphilis and other diseases.However,at present,the traditional cerebrospinal fluid cytology testing needs to be equipped with experienced pathologists to analyze a large number of images.The human cost is very high.Using artificial intelligence to replace the traditional cerebrospinal fluid cytology testing can not only greatly save medical resources,but also have important significance and application value in assisting clinical diagnosis and treatment.In practical application,on the one hand,the distribution of various types of cells in cerebrospinal fluid is seriously uneven,and the number of some important cells is very small;On the other hand,cells of the same class often show a variety of different forms due to the existence of different subcategories.Not only that,there is no obvious difference in visual characteristics between different types of cells in cerebrospinal fluid,which will seriously interfere with the recognition performance of the model.In view of the above challenges,this paper studies the methods of cell segmentation and classification of cerebrospinal fluid microscope images,proposes a cerebrospinal fluid cell detection and classification model with artificial features,and designs and implements an AI analysis system for cerebrospinal fluid cytology.The main contributions of this paper are as follows:1)In this paper,a CSF cell feature extraction model is designed.The cells are divided into three parts: nucleus,cytoplasm and background through u-net,and then the segmentation map is framed through Faster R-CNN.The required artificial features are obtained through the proposed cell feature extraction algorithm.According to the knowledge of cell morphology and geometry,this paper proposes seven artificial features of cell nucleus: roughness,concavity,aspect ratio,roundness,cell area,cell nucleus area and nucleoplasm ratio,and designs and implements the extraction algorithm.2)According to the obtained artificial features,a Faster R-CNN cell detection model with artificial features is proposed in this paper.The prediction results of the model are classified twice by using artificial features and machine learning algorithm,and the results of the cell detection model are interpreted by introducing s-life model to obtain the evaluation basis of the classification results.3)Starting from the actual application scenario,this paper designs the application scenario of the whole cerebrospinal fluid cytology analysis platform,and realizes the cerebrospinal fluid cytology AI analysis system.The cerebrospinal fluid cytology testing web system based on Springboot + Mybatis + Vue embeds the cerebrospinal fluid cytology testing classification model into the system,realizes the model-based prediction,classification result interpretation,cell counting,auxiliary labeling,supplementary training and other functions,and provides the whole process service from making cell smears to obtaining analysis results,which is of great significance to improve the hospital treatment efficiency and reduce medical costs.