WBC Subtype Detection using Deep Learning with Optimizing Hyperparameters by Genetic Algorithm
Mehedi Hasan Shuvo, Md. Tahamidur Rahman, Sk. Md. Masudul Ahsan · 2023
The human body's defense mechanism, the immune system, heavily relies on white blood cells (WBCs). These cells come in five subtypes: Eosinophil, Basophil, Neutrophils, Monocyte, and Lymphocyte. Each subtype plays a vital role in the diagnosis of hematological diseases, such as leukemia, as their presence or absence can indicate the presence of a disease. In the past, hematologists have manually examined blood smears under a microscope to detect the subtype of WBCs. However, this method is both time-consuming and prone to errors. This study proposes a new approach, which uses a deep learning-based convolutional neural network (CNN) architecture in combination with an evolution-based algorithm (genetic algorithm) to detect WBC subtypes with greater accuracy. A large dataset of 12,000 images was used for training and analysis. The proposed model, which utilizes the You Only Look Once version 5 (YOLOv5) for image localization and segmentation, was able to accurately localize WBCs with an IoU score of 94% for more than 80% of images and detect the subtypes with an accuracy of 98%. This study aims to overcome the limitations of previous research by providing a more efficient and accurate method for detecting WBC subtypes.