Fine-tuned YOLO-based deep learning model for detecting malaria parasites and leukocytes in thick smear images: A Tanzanian case study

Abstract

Malaria remains a serious public health concern in developing countries, where accurate diagnosis is critical for effective treatment. Reliable and timely detection of malaria parasites and leukocytes is essential for precise parasitemia quantification. However, manual identification is labor-intensive, time-consuming, and prone to diagnostic errors—particularly in resource-limited settings. To address this challenge, this paper proposes a fine-tuned deep learning model for detecting malaria parasites and leukocytes in thick smear images. The model is based on the YOLOv10 and YOLOv11 object detection architectures, each independently trained, validated, and evaluated on a custom-annotated dataset collected from hospitals in Tanzania to ensure contextual relevance. A fivefold cross-validation, followed by statistical analysis, was used to identify the best-performing model. Results demonstrate that the optimized YOLOv11m model achieved the highest performance, with a statistically significant improvement (p < .001), attaining a mean mAP@50 of 86.2 % ± 0.3 % and a mean recall of 78.5 % ± 0.2 %. These findings highlight the potential of the proposed model to enhance diagnostic accuracy, support effective parasitemia quantification, and ultimately reduce malaria-related mortality in resource-limited settings.