Artificial Intelligence–Supported Automated Microscopy for Malaria Diagnosis: A Multicenter Study

Abstract Background Accurate malaria diagnosis is key for patient management, surveillance, and control. Automated microscopy can overcome the variation observed among microscopists and is a promising new tool for diagnosis. The Noul miLab integrates smear preparation, staining, imaging, and artificial intelligence–supported parasite detection in a portable device. Methods A total of 2201 samples were collected from febrile patients across 2 sites in Ethiopia, where Plasmodium falciparum and Plasmodium vivax are frequent, and in Ghana, where P. falciparum transmission is intense. Samples were screened using local microscopy at the health center, miLab, and rapid diagnostic tests. Quantitative polymerase chain reaction (qPCR) and expert microscopy were used as gold standards. Results The miLab reached a sensitivity for P. falciparum diagnosis of 96.3% (335 of 348) using expert microscopy as the gold standard and a sensitivity of 97.4% (298 of 306) using qPCR-positive infections at densities >200 parasites/µL as the gold standard. The sensitivity of miLab for P. vivax was 96.8% (399 of 412) using expert microscopy as the gold standard and 95.9% (419 of 437) using qPCR-positive infections at densities >200 parasites/µL as the gold standard. Compared to qPCR, specificity was 98.8% for P. falciparum and 97.8% for P. vivax. miLab was significantly more sensitive than microscopy conducted at the health center. In Ethiopia, among miLab-positive samples, miLab assigned the correct parasite species to 98.7% of P. falciparum and 96.2% of P. vivax mono-infections. Conclusions The miLab automated microscope shows high sensitivity and specificity for P. falciparum and P. vivax diagnosis.