White blood cell classification using convolutional neural network

Abstract

The density of white blood cells in bloodstream provides a glimpse into the state of the immune system and any potential risks such as heart disease or infection. A dramatic change in the white blood cell count relative to your baseline is generally a sign that your body is currently being affected by an antigen. A variation in a specific type of white blood cell generally correlates with a specific type of antigen. Currently, a manual approach is followed for white blood cell classification; however, some semi-automated approaches have been proposed which involves manual feature extraction and selection and an automated classification using microscopic blood smear images. In this work, we propose deep learning methodology to automate the entire process using convolutional neural networks for a binary class with an accuracy of 96% as well as multiclass classification with an accuracy of 87%.