White blood cell automatic classification using deep learning and optimized quaternion hybrid moments

Abstract

The structural characteristics of white blood cells (WBCs) can provide important information about human health status. For this reason, over the past four decades, researchers have sought ways to automate the classification and morphological analysis of white blood cells. In the present study, we propose a new method to classify white blood cells into four types: neutrophils, lymphocytes, monocytes, and eosinophils by combining the hybrid discrete moment quaternion approach with deep learning. The proposed classification method is generally divided into two main phases: the first one is called the preprocessing phase, which is devoted to the computation of the image moments to be classified using a new quaternion Meixner-Charlier hybrid moment that is characterized by the following parameters α,β,andφ. In addition, the grey wolf optimization algorithm is used to guarantee high classification accuracy by optimizing the local parameters of the new quaternion Meixner-Charlier hybrid moments. The classification phase, which is the second phase of our proposed convolutional neural network model, is dedicated to introducing image moments. We present various graphical measurements, including receiver operating characteristics, precision-recall curves, and a confusion matrix. To demonstrate the effectiveness of our classification method, we also employ numerical measures such as the F1 score, loss, and accuracy. Our results indicate that the cell types of neutrophils were determined with a 98.53% accuracy rate, lymphocytes with a 98.16% accuracy rate, monocytes with a 97.43% accuracy rate, and eosinophils with a 97.43% accuracy rate.