Junhwi Park
|
Changjoon Park
|
Namjung Kim
and 2 other persons
| 2025, 30(1)
| pp.1~14
| number of Cited : 0
Blood is composed of white blood cells (WBC), red blood cells (RBC), and platelets, components closely related to human health. They respectively contribute to the immune system, the transfer of oxygen to body organs, and bleeding prevention. In the case of WBC, when abnormalities occur in the body, such as infections and allergic reactions, the relevant type of WBC in the blood increases to respond. Therefore, it is possible to identify the type of WBC in the case of abnormalities in the body through the application of deep learning-based object detection algorithms and classification methods.
This information can be used to predict the health status of patients easily. Therefore, this paper proposes a method for extracting the region of interest (RoI) for WBC and classifying WBC types based on blood microscope images using you-only-look-once (YOLO) and feature ensemble techniques.
We select RoI extraction models through a comparative analysis of WBC detection performance for YOLO V5, V8, V9 and YOLO-Neural Architecture Search (YOLO-NAS), and demonstrate an improvement in WBC type classification performance through comparative analysis and a Top-3 model feature ensemble based on general convolutional neural network (CNN) models, such as ResNet and EfficientNet. Compared to the approximately 98% performance of the Top-3 models based on the F1-Score, it achieved an improved performance of approximately 99%.