Peripheral White Blood Cell Classification Research Based On Neural Network Methods

Identification and classification of major white blood cell types named monocytes, lymphocytes, neutrophils, eosinophils, and basophils are important for clinical diagnosis.Automatic blood cell classification and recognition based on microscopic images can be assigned to the morphological detection category. It is one of the computer image processing and pattern recognition applications in clinical medicine. The more frequently utilized classification theory is statistics and artificial neural network. Neural network, especially fuzzy neural network is presented to classify the white blood cells.An adequate collection of digital color blood cell images is necessary. On the basis of the former research, combined with manual methods, cells are first segmented from the background. Then the nucleus and plasms are extracted respectively.Color characteristics, geometrical characteristics and texture characteristics are picked up from every single leukocyte image and stored in a certain form.After cell segmentation, leukocytes'feature acquirement and selection, a BP neural network classifier is constructed to determine the white blood cells'types. Especially, a classification method based on Sugeno-model fuzzy neural network is developed. The 3-types classification and 5-types classification experiment results show that the fuzzy neural network algorithm is more accurate for white blood cell classification.Furthermore, the relationship between feature acquirement or selection and white blood cell classification is discussed. Principal components analysis (PCA) is used to reduce the cells feature array. After the PCA, classification is researched again.