Label-free white blood cells classification using a deep feature fusion neural network

Area of Science:

• Biomedical Engineering
• Computational Biology
• Medical Imaging

Background:

• White blood cell (WBC) classification is crucial for disease diagnosis.
• Conventional methods like flow cytometry are costly and complex.
• Existing deep learning methods often overlook subtle intracellular features in WBCs.

Purpose of the Study:

• To develop a novel deep learning model for label-free WBC classification.
• To improve the accuracy of WBC identification by utilizing internal cell structures.
• To offer a simplified, cost-effective alternative to traditional diagnostic methods.

Main Methods:

• Proposed a neural network incorporating feature fusion.
• Combined low-level and high-level features from convolutional neural network (CNN) layers.
• Utilized bright-field microscopy images for label-free WBC detection.

Main Results:

• Achieved 80.3% accuracy on the testing dataset.
• Demonstrated effective utilization of subtle intracellular features.
• The method simplifies cell detection and eliminates staining requirements.

Conclusions:

• The proposed feature fusion network offers a promising approach for accurate, label-free WBC classification.
• This method has the potential to advance deep-learning-based biomedical diagnoses.
• It could facilitate the development of miniatured flow cytometers for point-of-care diagnostics.