Multiscale deformed attention networks for white blood cell detection

Area of Science:

• Medical Diagnostics
• Computer Vision
• Artificial Intelligence

Background:

• White blood cell (WBC) detection is critical for diagnosing various medical conditions.
• Traditional WBC detection methods are inefficient and time-consuming.
• Existing deep learning models like CNNs face challenges with global information and long-distance dependencies in WBC images.

Purpose of the Study:

• To develop an advanced deep learning model for accurate and efficient WBC detection.
• To overcome the limitations of current methods in handling large foreground-background differences in WBC images.

Main Methods:

• Introduction of the Multi-Scale Cross-Deformation Attention Fusion Network (MCDAF-Net), integrating Convolutional Neural Networks (CNNs) and Transformers.
• Development of the Attention Multi-scale Sensing Module (AMSM) for enhanced feature representation and accurate localization.
• Implementation of the Cross-Deformation Convolution Module (CDCM) to reduce feature correlation and improve generalization.

Main Results:

• MCDAF-Net demonstrates superior performance compared to existing models on public datasets (LISC, BCCD, WBCDD).
• The model effectively addresses challenges related to global information and long-distance dependencies in WBC image analysis.
• The fusion of CNNs and Transformers, along with specialized modules, leads to improved WBC detection accuracy.

Conclusions:

• MCDAF-Net offers a significant advancement in automated WBC detection.
• The proposed method holds promise for improving the speed and accuracy of medical diagnostics.
• The study provides a novel approach combining multi-scale feature fusion and attention mechanisms for cell detection.