High-Efficiency Classification of White Blood Cells Based on Object Detection

Area of Science:

• Medical diagnostics
• Computer vision
• Immunology

Background:

• White blood cell (WBC) analysis is crucial for diagnosing diseases.
• Traditional methods require cell segmentation, increasing workload and reducing efficiency.
• Automated WBC classification aids in early disease detection.

Purpose of the Study:

• To develop a high-efficiency object detection technology for simultaneous WBC detection and classification.
• To compare the performance of Faster RCNN and Yolov4 models for WBC classification.
• To evaluate the impact of eliminating cell pre-segmentation on classification quality and speed.

Main Methods:

• Employed deep transfer learning with Faster RCNN and Yolov4 object detection models.
• Utilized a balanced and enhanced Blood Cell Count Dataset (BCCD).
• Integrated segmentation and recognition into a single detection step.

Main Results:

• Achieved high classification accuracy rates: 96.25% for Faster RCNN and 95.75% for Yolov4.
• Yolov4 demonstrated a detection speed of 60 FPS with over 95% accuracy.
• The proposed method eliminated the need for cell pre-segmentation, significantly improving efficiency.

Conclusions:

• The integrated object detection approach enhances WBC classification efficiency and quality.
• Yolov4 offers a superior balance of speed and accuracy for real-time WBC analysis.
• This technology presents a viable solution for future point-of-care diagnostic systems.