Evaluation of an Artificial Intelligence-Based Tool and a Universal Low-Cost Robotized Microscope for the Automated Diagnosis of Malaria

Abstract

The gold standard diagnosis for malaria is the microscopic visualization of blood smears to identify Plasmodium parasites, although it is an expert-dependent technique and could trigger diagnostic errors. Artificial intelligence (AI) tools based on digital image analysis were postulated as a suitable supportive alternative for automated malaria diagnosis. A diagnostic evaluation of the iMAGING AI-based system was conducted in the reference laboratory of the International Health Unit Drassanes-Vall d'Hebron in Barcelona, Spain. iMAGING is an automated device for the diagnosis of malaria by using artificial intelligence image analysis tools and a robotized microscope. A total of 54 Giemsa-stained thick blood smear samples from travelers and migrants coming from endemic areas were employed and analyzed to determine the presence/absence of Plasmodium parasites. AI diagnostic results were compared with expert light microscopy gold standard method results. The AI system shows 81.25% sensitivity and 92.11% specificity when compared with the conventional light microscopy gold standard method. Overall, 48/54 (88.89%) samples were correctly identified [13/16 (81.25%) as positives and 35/38 (92.11%) as negatives]. The mean time of the AI system to determine a positive malaria diagnosis was 3 min and 48 s, with an average of 7.38 FoV analyzed per sample. Statistical analyses showed the Kappa Index = 0.721, demonstrating a satisfactory correlation between the gold standard diagnostic method and iMAGING results. The AI system demonstrated reliable results for malaria diagnosis in a reference laboratory in Barcelona. Validation in malaria-endemic regions will be the next step to evaluate its potential in resource-poor settings.

Keywords: Plasmodium; artificial intelligence; automated diagnosis; infectious diseases; malaria; point-of-care; tropical medicine.

Figure 1

Diagnostic flowchart of the iMAGING smartphone application software and ARIS software for Plasmodium parasites detection in thick blood smear samples.

Figure 2

Diagnosis accuracy results of the iMAGING system. Positive samples were divided by parasite densities (>10,000 parasites/µL, 801–10,000 parasites/µL, and <800 parasites/µL). Green crosses represent the mean number of Fields of View (FoV) observed for each sample category.