UR-Net: An Integrated ResUNet and Attention Based Image Enhancement and Classification Network for Stain-Free White...

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Differential white blood cell (WBC) counts offer vital disease insights.
• Traditional stained WBC classification is complex and sensitive to environmental factors.
• Stain-free WBCs present classification challenges due to inconspicuous nuclei, necessitating image enhancement.

Purpose of the Study:

• To develop an integrated model for enhanced stain-free WBC classification.
• To address limitations of standalone image enhancement techniques in computer vision tasks.
• To improve the accuracy of automated WBC classification from unstained samples.

Main Methods:

• Proposed a novel UR-Net model combining a ResUNet-based image enhancement network with an attention mechanism and a ResNet classification network.
• Integrated the enhancement module within the classification model for joint training.
• Utilized stain-free WBC images for training and evaluation.

Main Results:

• The proposed UR-Net model achieved a classification performance of 83.34% on a stain-free WBC dataset.
• Demonstrated superior performance compared to models without image enhancement and previous enhancement-classification models.
• The integrated approach effectively improved classification accuracy for challenging stain-free WBC images.

Conclusions:

• Joint training of image enhancement and classification networks significantly boosts performance for stain-free WBCs.
• UR-Net offers a promising solution for accurate and efficient automated WBC classification.
• This approach overcomes limitations of traditional methods and standalone enhancement techniques.