Enhancing White Blood Cell Detection Using YOLOv8 Model
Piyu Kasundra, Avani Khokhariya, Amit R Thakkar, Rohan Vaghela, Devanshi Dave, Jigar Subodhchandra Sarda · 2024
White blood cells (WBCs) play a vital role in the immune system, defending the body against pathogens such as viruses, bacteria, parasites, and fungi, and are produced in bone marrow and lymphoid tissues. Traditional WBC analysis methods rely on manual counting, which is prone to human error and inefficiencies. Recent advancements in computeraided detection systems, powered by deep learning, have significantly improved the accuracy and speed of WBC diagnostics. These systems utilize convolutional neural networks (CNNs) to automatically extract important features from blood smear images, minimizing reliance on manual techniques and enhancing diagnostic reliability. This study employed various YOLO v8 model variants to detect specific WBC types, with evaluation metrics including precision, recall, and mean Average Precision (mAP). The results demonstrates that the YOLOv8 medium and nano variants achieved the highest mAP50 (IoU threshold of 50) value of $\mathbf{9 9 \%}$, while the small variant achieved an mAP50 of $98 \%$ over 25 epochs. When comparing mAP50-95 (IoU threshold of 95), the medium variant attained $91 \%$, while the nano variant achieved $90 \%$. Additionally, the nano and small variants reached the highest average precision of $99 \%$, while the medium and nano variants recorded the highest average recall at $\mathbf{9 8 \%}$.