Classification of white blood cell from cellular hematology images

Abstract

This work presents a system for automated analysis of white blood cells in microscopic blood smear images for the detection and classification of leukemia subtypes. We constituted and preprocessed a data game of over 18 000 labeled white blood cell images covering four categories of leukemia as well as healthy samples. With the help of feature-refined convolutional neuronal networks, we obtain high accuracy to distinguish subtle morphological differences between these subtypes. The classifier is then integrated into desktop, web and mobile applications, offering real-time inference and a user-friendly interface for clinicians. The experimental results show the robustness of the system in the face of coloration variations and its interest for an early diagnosis.