Research On Algorithms Of Feature Extraction And Classification For White Blood Cells

As we all know,the number and percentage of various types of white blood cells(WBC)have great help for the diagnosis of medical diseases.Therefore,the study of WBC counting and classification has important applications.Because of the morphological changes of the same type of WBC,the technology of WBC classification has become a very challenging task.This dissertation studies algorithms for WBC segmentation and classification,including WBC segmentation based on location,WBC classification based on hierarchy,and WBC automatic recognition based on detection and deep learning.The main contents are as follows:1.According to the limitations of threshold segmentation algorithm Ostu,its segmentation result is relatively poor when the wave crest of histogram is not obvious,as well as PSNR.So we forward a segmentation algorithm based on the location,combining with the characteristics of WBC in the image.First of all,we locate the WBC using the information of nuclei and further dispose them with some morphological operations.After marking the approximate region where the white blood cell is,nuclei is segmented using Ostu algorithm in this region,which can reduce interference of background and other factors in segmentation.Finally the Grabcut algorithm is applied to segment cytoplasm.2.For the problem of automatic identification of WBC,this dissertation proposes a classification algorithm for its five types using their specific features based on the idea of hierarchical method.Firstly,we extract the leaflet feature and the circularity feature of the nuclei of white blood cells to screen the cells with these obvious characteristics.While for the cells without those obvious characteristics,we use the pairwise rotation invariant co-occurrence local binary pattern(PRICoLBP)feature to divide them into granular cells and nongranular cells.Finally,we divide the granular cells into basophils,eosinophils and neutrophils using PRICoLBP feature,and divide the nongranular cells into lymphocytes and monocytes using the ratio of the texture features and circularity features.Some experiments illustrate that our proposed method gets better performance than the existing corresponding hierarchical method.3.This dissertation proposes an automatic detection and classification system for WBC from microscope images based on detection and deep learning.Our proposed method firstly proposes an algorithm to detect WBC from the microscope images based on the simple relationship of colors and morphological operation.Then a granularity feature(PRICoLBP)and SVM are applied to classify eosinophil and basophil from other WBC.Last,convolution neural networks are used to extract features from the cropped WBC automatically,and a random forest is applied to recognize the other three types of WBC: Neutrophil,monocyte and lymphocyte.Some experiments on CellaVison database and ALL-IDB database show that our proposed method has better detection rate and recognition rate than some other methods with less cost time.