Algorithms for detecting and segmenting nucleated blood cells

The analysis of the different types of cells in blood is routinely used in today's medical practice to give an indication of a person's state of health. Many imaging systems and algorithms have been developed over the last 30 years in an attempt to automate this process. Some of these systems can now distinguish the difference between normal and abnormal cells but the differentiation among the various types of abnormal cells is still undergoing active research. A new system, the Cell Analyzer Imaging System, has been developed to acquire and process images from a microscope. In this work, some new algorithms have been developed using this system to detect and segment nucleated cells in Wright's stained blood smears for classification and sub-classification of the normal and abnormal cell types. The initial steps are to obtain high quality images by greatly reducing noise as well as by correcting distortions, aberrations and shading effects present in the acquired images. Spectral information from the images is then utilized to detect and segment nucleated cells from the rest of the scene (non-nucleated cells and background). All nucleated cells as well as those which are just touching are selected and separated into individual cells. The resulting single cells are further segmented into the regions of nucleus and cytoplasm. Simple features are then extracted from the segmented cells and these features are compared to determine if any clustering of a particular class of cell exists. Results show that these algorithms can detect, segment and classify different types of normal and abnormal nucleated blood cells. The major errors in segmentation accounts for approximately 6% of the cells analyzed.Applied Science, Faculty ofElectrical and Computer Engineering, Department ofGraduat