White blood cells detection using deep learning in healthcare applications
Abstract
Classifying and detecting the various types of White Blood Cells (WBCs) provides a valuable quantitative depiction of a body's health state and an important vision for the early treatment of illnesses. As a result, WBC classification and detection are critical. Traditional microscopic examinations require extensive time and complex procedures, thus reducing their reliable statistical results. The detection method for White Blood Cells requires automatic accuracy, therefore presents an important advantage. Nevertheless, the similarity of WBC samples and the insufficiency and imbalance of samples pose hurdles to automatic and accurate WBC categorization. This paper develops Op-YOLOv8 as an optimized YOLOv8 model designed to simultaneously analyze many white blood cell types while maintaining efficiency. The Op-YOLOv8 utilizes Maxpooling with depth-wise separable convolutions to compensate for feature loss through downsampling layers for maintaining vital contextual information. The proposed micro-scale detection layer uses a shallower feature map structure to deliver better YOLOv8 detection performance when dealing with overlapping and small-scale objects because it gathers complete contextual information. According to the BCCD dataset-based experimental results, the Op-YOLOv8 model performed better than other related models. Op-YOLOv8 demonstrates unrivaled WBC identification and detection performance through repeated validation of its maximum precision and recall alongside F1-score measurements reaching 0.981, 0.989 and 0.985 respectively for all monitored WBC types. The model's detection capabilities stand strong because it delivers outstanding mAP50 and mAP50–95 results, particularly for difficult classes Eosinophils and Neutrophils. The exceptional performance of Op-YOLOv8 stands out as it yields better precision and recall and F1-score metrics than related models. The model provides ideal performance for medical imaging scenarios requiring precise results. This model is a promising tool in medical imaging since it displayed high precision in different scenarios, indicating a great potential for application in various healthcare settings.