[A robust classification method for five types of leukocytes in peripheral blood based on mean-shift clustering]

Area of Science:

• Biomedical image analysis
• Hematology
• Computational biology

Context:

• Accurate classification of leukocytes (white blood cells) is crucial for diagnosing various diseases.
• Existing methods may struggle with subtle textural differences between cell types.
• Automated analysis of peripheral blood smears is a key area in diagnostic hematology.

Purpose:

• To develop and validate a new leukocyte classification method for five human peripheral blood smear types.
• To leverage mean-shift clustering for extracting visual texture features.
• To combine texture and geometric features for robust cell recognition using artificial neural networks.

Summary:

• A novel method extracts leukocyte texture features using mean-shift clustering and region growing.
• Feature points are identified and expanded to represent visual texture.
• A parameter vector derived from these regions, combined with geometric features, is used for classification.
• Artificial neural networks (ANN) are employed for recognizing five leukocyte types: neutrophil, eosinophil, basophil, lymphocyte, and monocyte.

Impact:

• Achieved high recognition accuracy rates: neutrophil (95.4%), eosinophil (93.8%), basophil (100%), lymphocyte (93.1%), and monocyte (92.4%).
• Demonstrates the feasibility and high robustness of the proposed automated leukocyte classification method.
• Potential to improve efficiency and accuracy in clinical hematology diagnostics.