On the Classification of Cytological Images of Leukocytes Using Depthwise Separable Convolutional Neural Networks

Blood cell analysis is the most important diagnostic process in medical practice. In particular, the detection of white blood cells (WBCs) is necessary for the diagnosis of many diseases. Manual screening of blood smears is labor-intensive and subjective, and can lead to inconsistencies and errors. However, automated blood cell detection can improve the accuracy and efficiency of the screening process. In this paper, a computer-based approach to the classification and detection of white blood cells in cytological images of blood cells using deep learning methods is implemented. A model for the classification of blood images, LeucoCyteNetv2, is proposed, which includes depthwise separable convolutional layers, Separable-Conv2D, in its architecture. The developed model classifies leukocytes with an accuracy of 98.86%, which made it possible to confirm its use as an auxiliary tool for hematological blood analysis. The proposed model represents a good alternative for automated diagnostics to support hematologists in the clinical laboratory in the evaluation of leukocytes in blood from blood smear images.