Automatic Method for Avian Red Blood Cell and Heterophil Counting Using Iterative Thresholding

Abstract

Thailand is one of the top broiler exporters. Red blood cell (RBC) and white blood cell (WBC) counting is one of the most basic methods for health screening process. Because the mammalian and avian blood is different, the mammalian blood analyzing techniques cannot be used, so the counting is done manually. This thesis proposes an automatic counting method for avian RBCs and heterophils, the most common type of avian WBC. The detection of avian RBC is challenging because of elliptic nucleated cell, the possibility of overlapped cells, and various staining. Otsu’s multiple thresholding method is used to automatically extract nuclei under different staining. The threshold value is then reduced iteratively to separate connected nuclei. Heterophil detection is done by removing the detected RBCs, using edge information to extract possible WBCs, then applying statistically based algorithm to filter heterophils. The results show that RBC counting error is less than clinical acceptable error of 5% for 4 out of 5 types of slide and heterophil counting error is 8.8%.