TW-YOLO: An Innovative Blood Cell Detection Model Based on Multi-Scale Feature Fusion

Area of Science:

• Medical Image Analysis
• Deep Learning
• Computational Pathology

Background:

• Automated medical image analysis is critical for clinical diagnosis, but traditional models struggle with the complexity of blood cell images.
• Existing methods show deficiencies in accurate blood cell detection due to image diversity and intricate cellular features.

Purpose of the Study:

• To develop an improved deep learning approach for enhanced blood cell detection in medical images.
• To address the limitations of traditional models in recognizing diverse and complex blood cell features.

Main Methods:

• Developed the TW-YOLO approach, integrating multi-scale feature fusion techniques.
• Incorporated Receptive Field Attention Convolution (RFAConv) to enhance geometric feature extraction.
• Utilized attention modules (CBAM, EMA) within a YOLO framework and introduced a PGI-Ghost strategy for refined gradient flow.

Main Results:

• TW-YOLO demonstrated superior performance in blood cell detection tasks.
• Achieved a 2% improvement over existing models on the BloodCell-Detection-Dataset (BCD).
• Showcased enhanced recognition capabilities for complex blood cell features.

Conclusions:

• The TW-YOLO approach offers significant advancements in automated blood cell image analysis.
• Provides robust technical support for future automated medical diagnostic systems.
• Highlights the potential of tailored deep learning architectures for specific medical imaging challenges.