Recognition of Abnormal Red Blood Cells Using Earth Mover's Distance Algorithm

Abstract

Red blood cells (RBCs) play a significant role by carrying essential nutrients and oxygen throughout the body. These cells are shaped as biconcave discs that are flexible enough to pass through small capillaries easily. However, if these cells are irregularly shaped, they may reduce the oxygen-carrying capacity of the blood and have difficulty moving through the blood vessels. Blood tests are laboratory methods that are often performed by medical experts to manually determine abnormalities in the blood, but it may be subjective and susceptible to human error. Therefore, recognizing abnormal RBCs automatically through digital image processing would be beneficial. There are existing studies that focus on the application of different image processing techniques in recognizing abnormal RBCs automatically. However, they suffer from certain weaknesses such as: others did not cover all the abnormal RBCs, while the others still encountered a significant error in recognition due to the slim dissimilarities between the attributes that they used. In this study, the proponents sought to improve and provide solution for the limitations of the previous studies by proposing a system that can recognize 9 abnormal RBCs using a machine learning algorithm called Earth Mover's Distance Algorithm which allows partial matching between two different distributions. The test images used in the system are microscopic images of blood samples gathered from the web. The images gathered initially undergone image preprocessing to extract the necessary attributes from multiple regions of interests (ROIs) needed in the computation of the Earth Mover's Distance (EMD). EMD is the measure of dissimilarity between two multi-dimensional distributions in space. The lowest computed EMD represents the smallest dissimilarity of the input image to the template image. As a result, the system displays the total number and label of each abnormal RBCs recognized in a Graphic User Interface (GUI). The proposed system obtained an average reliability rate of 93.52%.