Automatic White Blood Cell Classification Using Convolutional Neural Network
Tarza Hasan Abdullah, Fattah Alizadeh, Berivan Hasan Abdullah · 2023
Peripheral blood smear analysis is a commonly used technique for existing abnormalities in the health status of humans. Disorders in White Blood Cell (WBC) ratio imply the existence of diseases such as leukemia, lymphoma, anemia, and myelodysplastic syndrome (MDS). Deep learning techniques have gained promising results in many medical-related tasks such as dermatology, ophthalmology, and gastroenterology. In this paper, we developed a novel Convolutional Neural Network (CNN) architecture for leukocyte classification from microscopic peripheral blood cell images which help in the diagnosis of hematological disorders. To boost the performance of the proposed model and avoid overfitting phenomena we leveraged data augmentation techniques of rotation, flipping, width shift, and height shift. Unlike the conventional practice for designing a custom CNN model, we stacked homogenous layers and larger kernels in the early layers. To evaluate the performance of the proposed model, we have used classification metrics of accuracy F1- scores. The proposed model achieved an accuracy of 98.13% on the LISC dataset and then benchmarked against the state-of-the-art approaches.