Leucocyte classification for leukaemia detection using image processing techniques.

Area of Science:

• Medical image analysis
• Hematology
• Computational pathology

Background:

• Manual blood cell analysis is time-consuming and operator-dependent.
• Accurate white blood cell (WBC) classification is crucial for diagnosing diseases like acute lymphoblastic leukemia (ALL).
• Existing automated systems for blood cell analysis are often incomplete.

Purpose of the Study:

• To develop a fully automated method for WBC identification and classification using microscopic images.
• To compare different classification models for optimal ALL detection.
• To introduce a novel feature extraction approach for detailed cell component analysis.

Main Methods:

• A novel approach isolates the entire leukocyte before separating nucleus and cytoplasm for detailed analysis.
• Extraction of shape, color, and texture features using a new background pixel removal technique.
• Training and evaluation of various classification models, including support vector machines (SVM).

Main Results:

• The automated method achieved 92% accuracy in identifying 245 out of 267 leukocytes.
• Support vector machine with a Gaussian radial basis kernel showed the highest accuracy (93%) and sensitivity (98%) for ALL identification.
• The proposed feature set demonstrated superior performance across all evaluated classification models.

Conclusions:

• The developed automated image processing method offers a reliable alternative to manual blood cell analysis.
• This system can aid in early diagnostic suspicion for diseases like ALL.
• The method provides excellent performance for cell counting and classification, potentially improving diagnostic workflows.