A YOLOv8-based dual-validation study against nine experienced microscopists for detection of intraerythrocytic Plasmodium species and cellular mimics on Giemsa-stained thin films in an Indonesian endemic setting.

Microscopic examination of Giemsa stained thin blood films remains the reference standard for malaria diagnosis. However, accurate interpretation requires experienced microscopists and is complicated by staining artifacts and morphologically similar cellular structures. This study evaluated a YOLOv8 based morphology oriented detector for identifying intraerythrocytic Plasmodium species and differentiating cellular mimics in thin blood films from an Indonesian endemic setting. A dual validation diagnostic accuracy study used 4496 clinically verified microphotographs from 152 thin film slides for internal validation and 1701 images from 30 independent slides for external validation. Six object classes were annotated including four Plasmodium species, leukocytes, and uninfected erythrocytes. The detector was trained via transfer learning and evaluated against nine trained microscopists. Internal validation demonstrated macro averaged sensitivities and specificities of 80.25% and 88.19% respectively. Species level classification accuracy exceeded 99% for P. vivax, P. malariae, and P. ovale. External validation showed stable screening performance with 74.67% sensitivity and 88.25% specificity. The detector demonstrated improved sensitivity and precision compared with human microscopists during external testing and successfully differentiated intracellular parasites from morphological mimics. Morphology oriented automated analysis of Giemsa stained thin blood films demonstrated clinically useful capability for recognizing malaria parasites under heterogeneous endemic microscopy conditions. These findings support automated cellular morphology analysis as an adjunctive screening tool in low resource malaria endemic settings. However, significant limitations including dataset class imbalance, reduced sensitivity for early stage P. falciparum, and constrained external validation sample size underscore the necessity for broader multicenter evaluations prior to clinical deployment.