White blood cell type identification using multi-layer convolutional features with an extreme-learning machine

Abstract

White blood cells (WBCs) are the main component of the immune system that have a major role in protecting the body against different types of infections arising due to viruses, bacteria, fungi, and so on. The WBCs are generally referred to as 5 main subtypes: lymphocytes, monocytes, neutrophils, eosinophils, and basophils. Recognizing and counting each type of WBC is important for diagnosing and treating various disorders, such as infectious diseases, autoimmune disorders, immune deficiencies, leukemia, etc. To this end, a fast and accurate WBC classification model is crucial. This study offers a new model that works with a deep neural network—namely, multi-layer (ML) convolutional features of the AlexNet architecture followed by a feature selection (FS) strategy (MLANet-FS) for WBC-type identification. The proposed model exploits multi-layer convolutional features from different layers of the AlexNet model to provide rich discriminative detail, because different convolutional layers contain different visual characteristics of WBCs, and thereafter, linear fusion of these features occurs automatically. FS strategy is used to select the most distinguishing features from the feature fusion pool. Next, an extreme-learning machine (ELM) is employed to learn a discriminative model of WBC type identification. The proposed MLANet-FS-ELM model was evaluated in extensive experiments on the WBC benchmark dataset. It achieved 99.99% training accuracy and 99.12% testing accuracy, demonstrating that the proposed model outperforms alternative methods in the literature developed for WBC identification.