White blood cells (WBCs) are extensively dispersed in human blood, lymph, and other tissues, and they play a vital part in the human body’s immunological function. The prevalence of blood disorders associated with WBCs, such as leukemia and malignant lymphoma, is on the rise.

Red blood cells (RBCs), white blood cells (WBCs), and platelets make up human blood. WBCs, also known as leukocytes, are divided into two types: Non-granulocytes and granulocytes

The correct WBC count gives useful information for blood illness diagnosis, and it has become a major topic of research in clinical applications.

  • The BCCD Dataset is a small-scale dataset for detecting blood cells

WBC categorization

WBC categorization may be broken down into three types of techniques: microscopy, flow cytometry, and machine learning (ML).

The microscopy technique has the advantage of high precision; nevertheless, it has drawbacks such as low efficiency and a lengthy cell culture cycle.

Flow cytometry is now a common technique for quantitative analysis of blood samples, however, it is hard to conduct retrospective research of WBCs owing to the unavoidable destruction of blood samples.

Since ML has shown its potential for simplicity, repeatability, and robustness in WBC classification, it has been the most often used technique in practice.

ML algorithms for WBC classification

There are three types of ML algorithms for WBC classification:

  • conventional ML algorithms,
  • deep learning
  • hybrid techniques of machine learning and deep learning.

WBC morphological characteristics, according to standard ML algorithms, had a significant influence on WBC classification accuracy. The classifier of support vector machine (SVM) was coupled with the mixed characteristics of form, intensity, and texture, and the classification accuracy of 140 digital blood smear pictures of five types of WBCs could reach 84 percent. Furthermore, another WBC classification method based on synthetic features and random forest (RF) was developed for the categorization of 800 WBC pictures of five different types of WBCs, with a 95.4 percent accuracy. Furthermore, more cell energy and color characteristics were used in the K-means classifier to accurately identify 98 complete cancer pictures from the acute lymphoblastic leukemia image database for image processing (ALL-IDB).

DL has recently been popular in WBC categorization. Convolutional neural networks (CNNs) were utilized to categorize two kinds of WBCs from the ALL-IDB data set, with a 96 percent accuracy.

The hybrid technique of the combined CNN and recurrent neural network (RNN) model was used to identify four kinds of WBCs and achieved an accuracy of 90% for the BCCD data set. Furthermore, using the ALL-IDB data set, CNN categorization of five kinds of WBCs could achieve an accuracy of 96 percent.

Hybrid techniques were also utilized in the categorization of WBCs to improve performance. Multiple granularities and CNN characteristics were coupled with SVM and RF classifiers to classify five kinds of WBCs on mixed Cellavison, ALL-IDB, and Jiashan datasets, with an accuracy of 92percent.

DL outperformed two other MLs in terms of pipeline automation, translation invariance, weight sharing, and end-to-end training with improved parameter optimization. However, the DL algorithms performed differently for different data sets in terms of size and picture quality. Due to a lack of gold standard data sets and even execution of the suggested methods, the performance of DL algorithms has yet to be fully addressed. At the moment, there is no agreement on the strength of DL for the classification of WBCs, despite the fact that it has sparked a lot of interest in the field of WBC classification.