Deep learning based semantic segmentation of leukemia effected white blood cell.

Area of Science:

• Medical image analysis
• Computational pathology
• Biomedical engineering

Background:

• Accurate segmentation of white blood cells (WBCs) is crucial for diagnosing various diseases.
• Challenges in WBC segmentation include overlapping cells, variations in size/shape, and immature cell borders.
• Existing methods often struggle with the complexity of blood smear images.

Purpose of the Study:

• To develop and evaluate a novel, robust method for segmenting WBCs from blood smear images.
• To enhance the accuracy and reliability of automated blood cell analysis.
• To integrate advanced deep learning and image processing techniques for improved diagnostic imaging.

Main Methods:

• A hybrid approach combining UNet++ for pre-segmentation, marker watershed algorithm for separation, and Neural Ordinary Differential Equations (ODE) for refining segmentation.
• UNet++ generates probabilistic grayscale images, followed by marker watershed to resolve overlapping cells.
• ODE is applied post-convolution to minimize error propagation during training and inference.

Main Results:

• The proposed method achieved high segmentation accuracy with a mean intersection over union (Jaccard index) of 97.73%, Dice similarity coefficient of 98.36%, and mean pixel accuracy of 98.97%.
• The unique combination of UNet++, marker watershed, and ODE demonstrated superior performance compared to existing systems.
• The method effectively differentiates WBCs from Red Blood Cells (RBCs) and platelets based on structural variations.

Conclusions:

• The integrated UNet++, marker watershed, and ODE approach offers a significant advancement in automated WBC segmentation.
• This technique holds promise for enhancing clinical applications in automated blood cell analysis, diagnostic imaging, and disease monitoring.
• Future work should focus on optimizing the model for deployment on low-resource devices for point-of-care diagnostics.