LeukocyteMask: An automated localization and segmentation method for leukocyte in blood smear images using deep...

Area of Science:

• Digital pathology and medical image analysis.
• Computational biology and machine learning applications in hematology.

Background:

• Accurate identification and segmentation of leukocytes (white blood cells, WBCs) in blood smear images are crucial for diagnosing diseases like leukemia and AIDS.
• Challenges include variations in cell appearance, adhesion between cells, and image impurities, hindering robust analysis.

Purpose of the Study:

• To propose an end-to-end method, LeukocyteMask, for accurate leukocyte localization and segmentation in digital microscope images.
• To leverage pixel-level prior information for training deep convolutional neural networks (CNNs).

Main Methods:

• Utilized pixel-level prior information to train a deep convolutional neural network (CNN).
• Employed the trained CNN for leukocyte region of interest (ROI) localization.
• Obtained precise leukocyte segmentation masks via network forward propagation based on extracted ROIs.

Main Results:

• Experimental results demonstrate the effectiveness of the LeukocyteMask method.
• Quantitative and qualitative comparisons show LeukocyteMask achieves state-of-the-art performance.
• The method exhibits high robustness and accuracy in leukocyte segmentation.

Conclusions:

• LeukocyteMask provides a robust and accurate solution for leukocyte segmentation in digital pathology.
• The proposed deep learning approach effectively addresses challenges in blood smear image analysis.
• This method holds significant potential for improving disease diagnosis through automated cell analysis.