LWF-YOLO: a lightweight framework based YOLO for blood cell detection.

Area of Science:

• Medical imaging
• Computer vision
• Hematology

Background:

• Accurate blood cell detection is crucial for diagnosing hematological disorders like anemia and leukemia.
• Traditional methods face limitations in subjectivity and efficiency for complex cellular patterns.

Purpose of the Study:

• To introduce LWF-YOLO, a lightweight object detection framework for enhanced blood cell detection.
• To improve the accuracy and efficiency of automated hematological analysis.

Main Methods:

• Developed LWF-YOLO based on YOLOv11, incorporating a Multi-Scale Edge-Aware Feature Enhancement (MSEFE) module for cell boundary delineation.
• Introduced a Dynamic Channel-Mixing Gated Former (DCGFormer) and Dynamic Deformable Convolution Network (DyDCN) for optimized feature modulation and robust localization.
• Utilized Sobel operators, multi-scale fusion, identity mapping, dynamic channel splitting, adaptive gating, and multi-dimensional attention.

Main Results:

• LWF-YOLO achieved high mean average precision (mAP) in blood cell detection.
• Demonstrated competitive or superior performance compared to state-of-the-art models like YOLOv11 and RT-DETR-R18.
• Maintained a lightweight profile with minimal computational overhead, enabling efficient deployment.

Conclusions:

• LWF-YOLO offers a significant advancement in automated hematological analysis.
• The framework enables real-time diagnostics on resource-constrained devices, improving accessibility and speed.
• LWF-YOLO addresses the limitations of traditional methods by providing accurate and efficient blood cell detection.