Evaluation of an automated microscope using machine learning for the detection of malaria in travelers returned to the UK

Charles B. Delahunt, Roxanne R. Rees-Channer

Abstract:

Introduction: Light microscopy remains a standard method for detection of
malaria parasites in clinical cases but training to expert level requires
considerable time. Moreover, excessive workflow causes fatigue and can
impact performance. An automated microscopy tool could aid in clinics with
limited access to highly skilled microscopists, where case numbers are excessive,
or in multi-site studies where consistency is essential. The EasyScan GO is an
automated scanning microscope combined with machine learning software
designed to detect malaria parasites in field-prepared Giemsa-stained blood
films. This study evaluates the ability of the EasyScan GO to detect, quantify and
identify the species of parasite present in blood films compared with expert light
microscopy.
Methods: Travelers returning to the UK and testing positive for malaria were
screened for eligibility and enrolled. Blood samples from enrolled participants
were used to make Giemsa-stained smears assessed by expert light microscopy
and the EasyScan GO to determine parasite density and species. Blood samples
were also assessed by PCR to confirm parasite density and species present and
resolve discrepancy between manual microscopy and the EasyScan GO.
Results: When compared to light microscopy, the EasyScan GO exhibited a
sensitivity of 88% (95% CI: 80-93%) and a specificity of 89% (95% CI: 87-91%). Of
the 99 samples labelled positive by both, manual microscopy identified 87 as
Plasmodium falciparum (Pf) and 12 as non-Pf. The EasyScan GO correctly
reported Pf for 86 of the 87 Pf samples and non-Pf for 11 of 12 non-Pf
samples. However, it failed to distinguish between non-Pf species, reporting all
as P. vivax. The EasyScan GO calculated parasite densities were within +/-25% of
light microscopy densities for 33% of samples between 200 and 2000 p/μL,
falling short of WHO level 1 (expert) manual microscopy competency (50% of
samples should be within +/-25% of the true parasitemia).
Discussion: This study shows that the EasyScan GO can be proficient in
detecting malaria parasites in Giemsa-stained blood films relative to expert
light microscopy and accurately distinguish between Pf and non-Pf species.
Performance at low parasite densities, distinguishing between non-Pf species
and accurate quantitation of parasitemia require further development and
evaluation.