Gated Attention-Augmented Double U-Net for White Blood Cell Segmentation.

Area of Science:

• Medical Imaging Analysis
• Computational Biology
• Deep Learning

Background:

• Accurate white blood cell segmentation is crucial for disease diagnosis and monitoring.
• Existing U-Net based methods struggle with cell boundary ambiguity and low contrast.
• A need exists for robust segmentation models that handle complex cellular structures.

Purpose of the Study:

• To develop an advanced deep learning architecture for enhanced white blood cell segmentation.
• To address limitations in current segmentation techniques, particularly concerning subtle boundaries and overlapping cells.
• To improve the accuracy and robustness of automated white blood cell identification in medical images.

Main Methods:

• Proposed GAAD-U-Net architecture integrating attention-augmented convolutions and a gating mechanism.
• Utilized a Double U-Net base architecture for enhanced feature extraction.
• Evaluated the model on benchmark white blood cell datasets, including SegPC-2021.

Main Results:

• Achieved state-of-the-art performance on white blood cell segmentation tasks.
• Demonstrated a 3.4% Dice Score Coefficient (DSC) improvement on the SegPC-2021 dataset.
• Showcased superior segmentation accuracy, measured by Intersection over Union (IoU) and DSC, even on difficult images.

Conclusions:

• GAAD-U-Net effectively captures ambiguous boundaries and complex cellular structures.
• The proposed model offers a significant advancement in white blood cell segmentation accuracy and robustness.
• This work provides a powerful tool for medical image analysis and clinical applications.