Improved leukocyte classification in bone marrow cytology using convolutional neural network with contrast...

Area of Science:

• Immunology and Hematology
• Medical Image Analysis
• Computational Biology

Background:

• Leukocytes, or white blood cells (WBCs), are crucial for immune function and identifying various infections.
• Accurate classification of WBC types is vital for diagnosing conditions like leukemia and immune disorders.
• Traditional microscopy methods for WBC identification are labor-intensive and subject to observer variability.

Purpose of the Study:

• To develop a rapid and precise automated model for classifying leukocytes (WBCs).
• To enhance the accuracy and efficiency of WBC identification compared to conventional methods.
• To assist hematologists and pathologists in diagnosing blood disorders more effectively.

Main Methods:

• Utilized a large dataset of leukocyte images for training and testing a classification model.
• Employed transfer learning to accelerate the model training process.
• Applied Contrast Limited Adaptive Histogram Equalization (CLAHE) to improve image quality and model performance.

Main Results:

• The proposed leukocyte classification model achieved an initial accuracy of 81%.
• Integration of CLAHE significantly boosted the model's overall accuracy to 96.5% (a 15.5% improvement).
• The enhanced model outperformed existing state-of-the-art methods for leukocyte classification.

Conclusions:

• Image contrast enhancement techniques, specifically CLAHE, demonstrably improve Convolutional Neural Network (CNN) model performance.
• The developed automated WBC classification model offers a significant advancement over traditional diagnostic methods.
• This technology can substantially aid in the early detection of blood disorders and inform treatment strategies.