[White blood segmentation based on dual path and atrous spatial pyramid pooling]

Area of Science:

• Medical Imaging
• Computational Biology
• Hematology

Background:

• Accurate white blood cell (leukocyte) counting and recognition in blood smears are crucial for diagnosing hematological diseases, including leukemia.
• Manual analysis is prone to errors and variability, necessitating automated systems for reliable auxiliary diagnosis.
• Blood leukocyte segmentation is a foundational step for developing automated analysis systems.

Purpose of the Study:

• To develop and validate an advanced automated system for precise blood leukocyte segmentation.
• To improve the accuracy and efficiency of white blood cell analysis for clinical diagnosis.
• To enhance the capabilities of automated diagnostic tools for blood-related diseases.

Main Methods:

• An improved U-Net model incorporating a dual-path network in the feature encoder to capture multi-scale leukocyte features.
• Integration of atrous spatial pyramid pooling to bolster the network's feature extraction capabilities.
• A feature decoder utilizing convolution and deconvolution to reconstruct segmented targets to original image dimensions for pixel-level segmentation.

Main Results:

• The proposed algorithm demonstrated superior segmentation performance compared to existing methods on three distinct leukocyte datasets.
• Achieved a mean Intersection over Union (mIoU) value exceeding 0.97, indicating high segmentation accuracy.
• Qualitative and quantitative experiments confirmed the algorithm's effectiveness and robustness.

Conclusions:

• The developed dual-path U-Net model with atrous spatial pyramid pooling offers a highly effective solution for blood leukocyte segmentation.
• This method shows significant potential for advancing automated auxiliary diagnostic systems in hematology.
• The improved segmentation accuracy can contribute to more reliable and efficient diagnosis of blood diseases.