DefinePK

Abstract

In blood tests analysis identification of Red Blood Cell (RBC) morphology and count the RBC number is crucial to diagnose any symptoms of blood related disease. In current practice, such procedure is executed manually by a pathologist under light microscope. As the samples increased, manual inspection become laborious to the pathologist and since visual inspection is subjective, it might lead to variation to the assessed samples. To overcome such a problem, an automatic method is proposed by utilizing image processing procedure. Initially RBC regions are extracted from the background by using a global threshold method applied on a green channel color image. Next, noise and holes in the RBCs are abolished by utilizing a morphological filter and connected component labeling. Following that, geometrical information of the RBCs’ area is extracted to determine single and overlapping RBC region. The former region is further process to identify its morphology either normal or abnormal by using geometrical properties and Artificial Neural Network (ANN), while the latter will undergo cell estimation stage by using Circle Hough Transform (CHT) to estimate the number of individual cells. The proposed method has been tested on blood cell images and demonstrates a reliable and effective system for classifying normal/abnormal RBC and counting the RBC number by considering an overlapping constraint.