Segmentation and Classification of White Blood Cells Using the UNet.

Area of Science:

• Medical image analysis
• Computational pathology
• Hematology

Background:

• Accurate segmentation of white blood cells (WBCs) is crucial for diagnosing blood diseases.
• Existing image processing methods face challenges in blood cell segmentation due to image quality enhancement steps.

Purpose of the Study:

• To propose a novel model for accurate segmentation of leukocytes from blood samples.
• To evaluate the model's performance against existing approaches for hematology data analysis.

Main Methods:

• A novel model combining ResNet and UNet networks was developed for feature extraction and segmentation.
• The model was evaluated using three distinct WBC datasets and cross-validation.
• Performance was assessed against established methods like DeepLabV3+ and ResNet-50.

Main Results:

• The proposed model achieved an overall segmentation accuracy of approximately 96%.
• The model demonstrated superior performance compared to DeepLabV3+ and ResNet-50.
• Experimental results indicate the model's suitability for analyzing hematology data.

Conclusions:

• The developed ResNet and UNet based model offers a highly accurate tool for WBC segmentation.
• This approach shows significant potential for improving early and accurate identification of blood diseases.
• The model's high accuracy suggests it is a valuable advancement in automated hematology analysis.