White Blood Cell Image Classification For Assisting Pathologist Using Deep Machine Learning: The Comparative Approach
A. REDE KIRTEE · i-manager’s Journal on Image Processing · 2020
White Blood Cell (WBC) is known as leukocyte or white corpuscle. It defends the body against infection and disease by removing foreign materials and cellular debris. The normal range of WBC in healthy people varies from 4,000 to 11,000 WBCs per microliter. If this count lowers, it indicates various diseases, such as viral infection, cancer, leukemia, etc. and high count indicate diseases such as infection, stress, allergies, etc. Traditional manual inspection of blood smear image is replaced by computerized technology and currently with the help of deep learning approach, we can extract accurate count and detailed condition of WBCs, which helps pathologist in terms of less labour and less time. We use WBC dataset having 12,500 images for experimentation on deep convolutional neural networks like AlexNet, VGG16, VGG19, InceptionV3, and ResNet50. To get better result, each network is modified by changing the layers in original architecture. Further various hyper parameters, such as learning rate, regularization, training epochs, batch sizes, etc., are adjusted to improve accuracy. All these networks have been compared for selection of appropriate network. InceptionV3 gives highest accuracy (99.6%) for WBC classification.