Blood Cell Counting Using Watershed Algorithm

Abstract

Blood cell counting is traditionally carried out either manually with a haemocytometer or through automated 
analyzers, and both approaches have drawbacks. Manual counting is slow and laborious, and errors creep in because 
cells overlap and visual inspection is not consistent from one observer to the next. Automated analyzers are expensive 
and cannot detect variations in cell shape or other irregularities. This paper proposes a digital image processing 
method that improves accuracy while reducing the time and cost of blood cell analysis. The process begins with 
acquisition of a stained blood sample image using a microscope and camera. The image is then pre-processed to 
remove noise and to extract the colour planes corresponding to red blood cells and white blood cells. Cells are 
separated using the watershed algorithm, which treats pixel intensity as elevation and floods the image from marker 
points so that touching and overlapping cells are split into distinct regions. Counting is then performed on the 
segmented regions using shape and size criteria. The method achieved an accuracy of 94% for red blood cell counts 
and 92% for white blood cell counts when compared with manual counting. A MATLAB application module was also 
developed, allowing a user to upload an image and receive the cell counts immediately.