Abstract
BACKGROUND: Malaria diagnosis plays a key role in case management, control, and elimination strategies. miLab™ is a digital microscopy with a fully integrated, sample-to-result approach, providing automated microscopic analysis of Plasmodium parasites and providing parasitemia levels of samples. It uses a deep learning model, a subfield of artificial intelligence (AI) that can differentiate from red blood cells that are infected with the malaria parasite from noninfected cells in blood smears. The aim of this study is to assess the performance of miLab™ microscopy for malaria diagnosis, in comparison with conventional microscopy and nested-multiplex malaria polymerase chain reaction (NM-PCR), in a malaria reference laboratory in a nonendemic country. METHODS: From 2021 to 2024, 400 samples were analyzed prospectively using automated miLab™ microscopy, with NM-PCR and conventional microscopy as reference methods. RESULTS: The comparison between the miLab™ device and thin blood smear microscopy showed substantial concordance (90.8%), with a kappa coefficient of 0.8 and sensitivity and specificity values of 92.1% and 89.4%, respectively. The comparison of parasite density showed a significant correlation (correlation coefficient of 0.77), although the parasite counts estimated by the miLab™ device were 11.6% lower than those estimated by conventional microscopy. The sensitivity and specificity values of the miLab™ platform when compared with those obtained by NM-PCR were 62.8% and 95.4%, respectively; with a concordance value of 68.9% (kappa coefficient 0.4). Of P. falciparum infections identified by NM-PCR, 63.4% were accurately identified, and this figure increased to 95.7% if excluding negative results. One P. vivax, three P. ovale, and one P. malariae infections identified by NM-PCR were correctly classified by the miLab™ platform only after expert review of initial "review needed" results. CONCLUSIONS: miLab™ automated microscopy was as sensitive as conventional microscopy but without the need for expert microscopists and with shorter time to results. It is a valuable toolkit for malaria diagnosis in nonendemic settings; however, improvements are required in terms of species identification and parasite quantification.