Research On The Recognition System Of Urinary Sediment Based On Neural Network And Fuzzy Inference

Biomedical image processing is a new technology and it is also an important direction of image processing. With the development of biomedical technology, the application of the biomedical image processing is more and more extensive. Urinary sediment microscopic inspecting is an important method in clinical inspection. The traditional optic-microscopical inspection which is watched by one eye is easy to not only increase doctors' labour intensity but also bring up artificial error. It is impossible to get standardized. Processing and analysis of the urinary sediment microscopical images based on computer system can increase efficiency of clinical inspection greatly, relieve the burden of doctors in clinical laboratory, provide standardization of disease diagnosis and manage present & history information of clinic. Moreover, it is easy to share medical datum and realize long-range consultation. This paper summarized the current research progress of urinary sediment image processing and components recognition at home and abroad. We do many researches in identifying the components in urinary sediment images such as image enhancement, image segmentation, feature extraction and classifier design. Through amount of theoretical researching and experiments, a set of complete effective processing scheme for urinary sediment visible component is proposed by combining multi-processing arithmetic from the characteristic of urinary sediment images.It is necessary to correctly segment image in order to analyze accurately urinary sediment visible component. We found that self-adaptable wiener filter has a better effect in preserving edge information than median filter by contrast experiments on image pre-processing stage. The traditional segmentation algorithm has been improved in order to gain better result. High-microscopical images segmentation are based on canny edge detection. Red blood cell, white blood cell and kinds of crystals can be successfully segmented out from the background. Low-microscopical images segmentation are based on two-dimensional OTSU adaptive thresholding. Epithelial cells and casts can be successfully segmented out from the background.Three-layer BP neural network model was applied to recognition of visible components. In order to improve the recognizing rate of the components, fuzzy inference system was introduced after neural network. The recognition experiences of experts were transformed into 'if-then' rules. Mamdani fuzzy inference system which has multi-rules and multi-inputs was created. Good results were obtained in classification of white blood cell and epithelial cells by fuzzy inference. Experiment results prove that fuzzy inference system can improve the recognizing rates of the abnormal components. The accuracy and recognizing rates of the whole system can be greatly improved by the combination of neural network, and fuzzy inference.