WBCaps: A Capsule Architecture-based Classification Model Designed for White Blood Cells Identification.

Area of Science:

• Hematology
• Medical Image Analysis
• Computational Pathology

Background:

• Accurate identification of white blood cells (WBCs) is crucial for diagnosing various hematological conditions.
• Manual microscopic examination of blood smears is time-consuming and prone to inter-observer variability.
• Automated methods are needed to improve the efficiency and consistency of WBC classification.

Purpose of the Study:

• To develop and validate an automated classification model, WBCaps, for recognizing five types of WBCs from peripheral blood smears.
• To present a complete workflow for WBC identification, including cell segmentation and classification.
• To compare the performance of WBCaps against established deep learning models like ResNet and VGG.

Main Methods:

• A three-step image segmentation method (color normalization, deconvolution, cell extraction) was used to locate WBCs.
• A capsule architecture-based model, WBCaps, comprising convolutional, primary capsule, and representation capsule layers, was developed for classification.
• 3-fold cross-validation was employed to evaluate the model on a clinical dataset.

Main Results:

• The WBCaps model achieved high performance metrics: precision of 0.99, recall of 0.99, and F1-score of 0.99.
• The proposed model significantly outperformed ResNet (precision 0.97, recall 0.97, F1-score 0.97) and VGG (precision 0.97, recall 0.97, F1-score 0.98).
• The segmentation method effectively extracted WBCs for subsequent classification.

Conclusions:

• WBCaps demonstrates a highly accurate and robust automated method for WBC classification from blood smears.
• The model's superior performance suggests its potential as a valuable tool for hematology analyzers.
• This automated approach can facilitate cytological and morphological examination, aiding in clinical diagnostics.