Detection and Identification of RBC Abnormalities in Essential Thrombocythemia Patients Using ResNet50V2

The MPN condition has been observed to present with several abnormalities in RBCs, including an increased RBC count and alterations in RBC shape. The elevated number of red blood cells (RBCs) in ET is responsible for the increased viscosity, which subsequently reduces the velocity of blood flow. The initial examination for early detection is a full blood count (FBC) but cannot exclude an increase in the number and variation of red blood cells (RBC). A limitation of the FBC examination is failure to detect cell morphological abnormalities, requiring instead microscopic examination of peripheral blood smears. Furthermore, microscopic examinations implemented in hospital laboratories are still performed manually by clinical pathologists. It is evident that the manual examination is inherently subjective, dependent on the clinical pathologist's experience. It is essential to develop a technological solution that enables the identification of RBC shape abnormalities through the application of computer vision to facilitate the work of clinical pathologists. The objective of this study is to apply the ResNet-50V2 algorithm to the classification of red blood cell (RBC) abnormalities in patients with essential thrombocythemia. The model effectively classified five classes of RBCs (normal erythrocytes, elliptocytes, ovalocytes, stomatocytes and teardrops) with the highest accuracy of 95% at a learning rate of 0.001 and a maximum iteration of 50 epochs.