The ability to make a successful morphological diagnosis of malaria remains an important and cost-effective health intervention worldwide [1]. However, malaria detection and recognition on blood films require individual skills, knowledge, training, and experience. The effects of misdiagnosis can be very serious [2]. Supporting the quality of diagnosis in different countries and settings can be challenging, particularly for those individuals working in isolation [3,4]. Direct practical instruction and education at the microscope are central to training [5], and glass-slide specimens of malaria-infected blood distributed by reference laboratories provide excellent education in diagnosis [6]. External quality assessment (EQA) has been shown to improve accuracy [7,8]; however, it has also been difficult to implement since the collection and provision of clinical materials are labor intensive, and slide numbers may be limited [9-11].

With facilities for Internet access improving globally, it is now possible to deliver digital slides of blood films to a wide range of geographical locations. Web-based virtual microscope systems [12,13] greatly support this process, and the skillsets have been shown to mirror those used in glass slide diagnosis [14]. The virtual microscope has been used in university education [15-18], EQA schemes [19], and for continuing professional development [20,21].

This study aims to demonstrate the feasibility and effectiveness of the virtual microscope to deliver digital slides of malaria-infected blood with high-quality resolution and to assess how the technique can be used to support skills in malaria diagnosis in developing and developed nations.

Participants for this study were recruited with help of the World Health Organization (WHO) and Liverpool School of Tropical Medicine. We used digital slides prepared from 56 cases that had previously been fully validated and distributed as glass slides through the UK National External Quality Assessment Scheme (UK NEQAS) parasitology scheme, confirmed by molecular techniques and consensus opinion of more than 400 laboratories. The cases were selected to represent different slide preparation techniques, stain quality, malaria species, and parasite density. Images were captured using a Zeiss Axio Imager M1 microscope with HRc camera and 63x Plan Apo Chromat 1.4 Oil immersion lens. At least 40 adjacent fields were acquired to create a single large stitched-image. Subsequent image processing used Zeiss software (Axiovision 4.7), then Photoshop CS3 to ensure sharpness and color balance. Diagnostic features were clearly resolved (contrast mask and detail enhancement using Digital Outback Photo add-in), and image size for upload was around 200 megabytes. Images were uploaded to the viewing software (Digital SlideBox, Leica Biosystems) and assessed to ensure the inclusion and accurate rendition of features required for diagnosis. Registered participants were from five countries (ie, Hong Kong, India, Kenya, Lebanon, and Nigeria), and some received financial support for Internet access. In all cases, the participants were asked to suggest their preferred diagnosis (multiple-choice options). Responses were analyzed using Microsoft Excel and GraphPad Prism software (v6.04).

The virtual microscope environment allowed low-power image scanning, navigation, and high-power view assessment (see Figure 1).

The images showed that the initial high-quality image capture clearly resolved those features required for malaria diagnosis, species assignment, or detection of artifact (see Figures 2 and 3). Images were then reassessed following assembly into a large single slide, and finally as a screen capture of a compressed Web-displayed image. Comparison between initial image and Web-browser images revealed an expected subjective loss of definition, but diagnostic detail was retained (see Figure 4).

The digital slides were then viewed by 38 participants from a range of countries: Hong Kong (n=1), India (n=1), Kenya (n=7), Lebanon (n=6), and Nigeria (n=23). The training and viewing environments differed between groups, and facilities used in different countries were not identical. Although images were viewed predominantly at hospital sites, a significant number used Internet cafés (with financial support). Aspects of the viewing environment were uncontrolled (ie, screen resolution and Internet speed). Although positive experience was reported, the impact of these aspects on results was unclear. Those using Internet cafés had lower rates for sensitivity and specificity for their diagnosis. The mean completion rate was 75% (42/56 cases). Each individual case was completed by a mean of 29 participants (range 22-38). The outcome of slide analysis (indicated by the submitted diagnosis) was compared with parasite density and type. Results indicate that successful recognition and species identification, in keeping with known results of glass slide analysis, were closely linked with the number of malarial parasites present on the slides (see Figure 5). Thick and thin films were assessed with equivalent success, and films containing no parasites were effectively identified. Similarly, in keeping with recognized findings, the recognition that a parasite was Plasmodium falciparum or non- P. falciparum species was good. However, the precise species-identification for non- P. falciparum species was less accurate (see Figure 6).

Finally, results were assessed to see whether they could be used to allow participants to assess their individual performance, comparing their results with other participants with similar training or experience (see Figure 7). The performance data across a consecutive case series were also plotted and used to make a cumulative assessment that allowed individuals to compare their performance trend with other users (see Figure 8).