YOLOv8-Based Malaria Parasite Detection in Microscopic Blood-Smear Images

Abstract

Accurate and timely malaria diagnosis remains challenging due to the limitations of manual microscopy, which is labor-intensive, subjective, and heavily dependent on expert availability. To improve diagnostic reliability and automation, this study investigates the capability of the YOLOv8 object-detection framework to identify malaria Trophozoites and white blood cells (WBCs) in thin blood-smear images. A YOLOv8 model was trained on annotated smear images and evaluated using precision, recall, mAP50, and mAP50-95. Its performance was compared with a baseline CNN classifier to assess differences between image-level classification and localized detection. The proposed YOLOv8 system achieved strong detection performance, including a precision of 0.857, recall of 0.857, and mAP50 of 0.884, while successfully distinguishing between Trophozoite and WBC classes. These findings indicate that YOLOv8 provides accurate multi-class localization and represents a promising direction for automated malaria screening and computer-assisted microscopy.