Automatic Detection And Recognition System Research On Urinary Sediment Images

Automatic optical-image-based urine sediment analysis is aimed at solving theproblem on identification of visible components in microscopic images of urinary withimage processing and pattern recognition techniques. By statistics of the visiblecomponents, automatic urine analysis helps medical staff to diagnose the patient’scondition. In traditional urine sediment analysis, the medical staff needs to observe andcount the visible components with naked eyes. Thus, traditional optical-image-based urinesediment analysis is heavily depended on the experience of medical staff, and it’s veryinefficient and tedious. With application of the image processing and pattern recognitiontechniques, optical-image-based urine sediment analysis can be more automatic. However,the optical image of urine sediment is very complicated, so it makes accurate detection ofall kind of visible components very difficult.In this paper, different segmentation algorithm is applied to the detection of visiblecomponents in urine sediment images for comparison, and then, a final solution isproposed for the recognition of the visible components. As the accuracy of recognition ofvisible components is heavily depended on the result of segmentation of the urinesediment images, we have compared global-threshold-based, edge-based, region-basedsegmentation and level-set-based segment method on the urine sediment images. Then theedge-based segment is used in the segmentation of all visible components, andmarked-watershed and level-set-based segmentation is also used respectively aimed atdealing with the segmentation of the adherent cell and some weak-edge visiblecomponents. At last, decision tree is applied to distinguish blood cell, crystal cell, andepithelial cell with morphological characteristics, and SVM for red blood cells and whiteblood cells with texture feature.