Blood Cell Detection and Counting Using Bounding Circle Representation

Accurate detection and counting of blood cells are crucial for hematological analysis. Traditional microscopic examination is time-consuming and prone to variability. Deep learning, particularly YOLO-based models, has significantly improved real-time object detection. This study improves YOLOv11 for blood cell detection by introducing bounding circles instead of conventional rectangular or square bounding boxes. Bounding circles align better with the natural morphology of blood cells, reducing background noise, and improving localization accuracy. The proposed method is evaluated on publicly available blood cell datasets and compared with the standard YOLOv11 model using square bounding boxes. The experimental results show detection accuracies of 100% for red blood cells (RBC), 95. 08% for white blood cells (WBC), and 96. 36% for platelets. The mask detection ratio (MDT) analysis further confirms the effectiveness of the method, achieving an average MDT of 87.93%, compared to 76% for conventional bounding boxes. The advantages of using bounding circles for deep learning-based blood cell detection are demonstrated, offering a more morphology-aware alternative for improved accuracy.