Classification of Agranulocytes using Neural Network

V.R. Udupi, A.V. Deshpande, and H.P. Inamdar (India)




The leukocyte population in peripheral blood smear is normally constituted of cells with granules in their cytoplasm, called granulocytes and those without cytoplasmic granules, called non-granulocytes or agranulocytes. The existing practices for the identification of leukocytes require more calculations and needs more time. In this study, over 100 cells are pre-processed, segmented and analyzed for extracting various discriminative features such as nucleus size, nucleus shape, nucleus to cytoplasmic area ratio, etc. The classification of agranulocytes into three category viz. Lymphocytes, Monocytes and Neutrophils is performed using feed-forward artificial neural network trained using Back-propagation rule. The network converges for 10,000 training cycles. The network provides the overall success rates of 88 percent of accuracy in identifying the agranulocytes from the sample images.

Important Links:

Go Back