[A leukocyte pattern recognition based on feature fusion in multi-color space]

Area of Science:

• Medical image analysis
• Computational pathology
• Hematology

Context:

• Leukocyte classification is crucial for diagnosing blood disorders.
• Existing methods struggle with multi-feature fusion in single color spaces.
• Developing robust automated systems is essential for clinical efficiency.

Purpose:

• To develop an automatic leukocyte pattern recognition system using multi-feature fusion across multiple color spaces.
• To evaluate the performance of different color spaces, features, and distance metrics for leukocyte classification.
• To establish an optimized, integrated system for high-precision leukocyte identification.

Summary:

• The study proposes a novel method for leukocyte classification by fusing color histograms and texture granular features in multiple color spaces (RGB, HSV, Lab).
• It analyzes the interaction of these features with various similarity metrics (normalized intersection, Euclidean, chi2, Mahalanobis distance).
• An optimized, tree-integrated recognition system was developed, achieving significant performance improvements.

Impact:

• The fusion classification approach demonstrated at least a 12.3% performance improvement over existing methods.
• This enhanced system offers high precision, broad applicability, and cost-effectiveness for different leukocyte types.
• The findings pave the way for more accurate and efficient automated blood cell analysis in clinical settings.