Research On Automatic Detection And Image Processing For Urine Micro-particles

Urine microscopic particles detection is one of the three routine examinationitems in clinical medical inspection operating rules. The National Clinical TestRegulation of Operation of third edition had definite requirements for urinemicroscopic particles inspection. But in most hospitals the urine microscopic particlesonly could be inspected by the artificial microscopy method or partial automaticinstruments, which ended in low speed progress, poor measurement traceability andlow standardization, the inspection results was largely dependent on the personalexperience of clinical laboratory physician. Combined with the requirements of urinemicroscopic particles automatic inspection, the adding sample mechanism,intelligentcontrol, machine vision, image processing and pattern cognition of the inspectionsystem were studied in this dissertation.A fully automatic and multi-channel online detection system was designed on thepart of system configuration and detection method, with such main part as samplemechanism and fluid systems, the microscope automatic focusing system based onARM and system software. Considering the long-time natural sedimentation of oldmethods and spatial distribution characteristic of formal ingredient in urine, thespecial dynamic coordinates track detection method was put forward to improve thedetection speed and detection effect.As to the microscope imagine processing aspect, different factors affecting theimage resolution were analyzed and a highly efficient resolution evaluation functionwas designed, the multi-sensor microscope image automatic focusing system wasbuilt with image resolution as control objective, which gave the detection system highquality image continuously and steadily.The special multi-focal plane image fusion model was built and its modelalgorithm was put forward, also with the image fusion method based on wavelet andwavelet packet transform, the algorithm was verified by experiments and proved tohave good effects on eliminating fake positive specimen and ensuring validitydetection.The formal ingredient automatic segmentation of image is the core technology ofurine microscopic particles automatic detection, the efficiency of image segmentationalgorithm has direct influence on the validity of morphological characteristicsextraction. Various image segmentation algorithm and its effects were investigated inthe dissertation, and edge detection algorithm was designed based on dyadic wavelet and wavelet packet. The Pulse Coupled Neural Network (PCNN) segmentationalgorithm was introduced to deal with binding components in image, and theexperiments indicated that PCNN had good effects to extract binding componentsbetween cells.Detection accuracy in the main examination index of urine microscopic particlesdetection system, the feature extraction efficiency and performance of formalingredient directly affect the accuracy of pattern cognition. The image featureextraction method was studied on the basis of segmentation, also the urinary formalingredient was classified and counted by BP neural network with its features wereanalyzed from aspects of shape, structure, texture, frequency domain, particulardomain and etc, and the neural network was proved to have good classification effectsafter training and testing.The research results of this dissertation have been applied to use in FullyAutomatic Urine Microscopic Particles Analysis Instruments of Suzhou HyssenElectronic technology Ltd, thus to improve the automatic level of the urinemicroscopic particles detection and get widely used.