Citation: Gaunt MW, Crainey JL (2024) AI sees an end to filariasis. PLoS Negl Trop Dis 18(7): e0012260. https://doi.org/10.1371/journal.pntd.0012260

Editor: Shaden Kamhawi, National Institutes of Health, UNITED STATES

Published: July 11, 2024

Copyright: © 2024 Gaunt, Crainey. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Funding: JLC acknowledges support from Conselho Nacional de Desenvolvimento Científico e Tecnológico CNPq09/2023_PQ315260/2023-8 and Programa de Excelência em Pesquisa Básica e Aplicada em Saúde dos Laboratórios do Instituto Leônidas & Maria Deane for 2023-2025. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: The authors have declared that no competing interests exist.

The WHO has successfully controlled lymphatic filariasis and onchocerciasis transmission with mass drug administration (MDA) in almost all the areas it has targeted [1,2]. The success of Latin American filarial disease control programs means the region will soon be free of both onchocerciasis and lymphatic filariasis; however, the elimination of these diseases in Africa and the control and elimination of other filarial diseases, like loiasis and mansonellosis, will require blood surveys in regions with little or no base-line surveillance data [3–7]. Historical expansions of filarial disease interventions required significant investment in specialist training for the collection of reliable surveillance data and thus the new digital technology reported by Lin and colleagues in this journal is a potential game changer [7–9].

Imagine a future where training for 100% reliable filariasis diagnosis is as simple as downloading an app and connecting a cellphone to a microscope eyepiece. Lin and colleagues describe a computer vision platform which to our knowledge is unprecedented for its filariasis surveillance potential: it’s an app available on Google Play, does not need specialist equipment or sample preparation, and there are no obvious barriers to its rapid roll-out [9].

Wucheceria bancrofti and B. malayi are both current epidemiologic targets of WHO MDA interventions; however, the app’s ability to discriminate M. perstans and L. loa microfilariae is arguably of more interest to the WHO [6,7,9–11]. Almost all the WHO’s MDA in Africa is done with ivermectin doses that can be fatal for patients with heavy loiasis parasitic loads and thus close monitoring of loiasis infections is required in all ivermectin-based MDA operations [10–12]. There are currently around 186 million Africans infected with M. perstans and it is the most likely microfilariae to appear on blood smears and be misdiagnosed as L. loa in WHO intervention zones [13,14]. Thus, the app has significant epidemiological potential.

The Lin and colleagues [9] app is advanced tech that is evident by a de facto model for computer vision, convolution neural networks (CNN), used by the tech giants such as Facebook. CNN, for example, looks for differences at 2D “boundaries” then stores this “learning” to memory. The authors described their AI as showing good overall performance with species specific variation. The diagnostic species “precision”, which approximates to true positives versus false positives, was 95% to 99% for L. loa, 60% to 97% for M. perstans, 95% to 100% for W. bancrofti, and 59% to 67% for B. malayi. Thus, a high false positive rate is occurring for B. malayi diagnosis which is corrected by further AI training. In addition, the authors demonstrated the AI had high “recall” (“sensitivity”), approximating to true positives versus false negatives, with 92% to 100% for each species, excepting a value of 76% for W. bancrofti. Thus, if the AI identifies W. bancrofti, then that is a statistically significant observation, but will miss around one quarter of infections from this species. However, again deeper AI training will resolve the matter.

Under the hood, when the algorithm “sees” a thick-blood smear slide it has not encountered, it checks its “learning” to classify the unknown image and hence gives species diagnosis. Unlike human vision, small visual differences between filariae species are not a barrier to progress with CNN because it can readily turn “molehills” to “mountains” through “weighting” once the pattern is identified. Identifying different Mansonella microfilariae, while difficult for optical analysis, is unlikely to present a challenge to a future version of the app described by Lin and colleagues [9].

There is no immediate reason why the WHO would need an app capable of discriminating between M. perstans and M rodhainii in Africa, but an ability to distinguish between M. perstans from M. ozzardi infections in Latin America would be valuable [5]. Mansonella perstans is refractory to ivermectin therapy, while M. ozzardi is susceptible; thus, future mansonellosis control needs reliable diagnosis between these 2 parasites for intervention, planning, and monitoring [5,14].

Lin and colleagues [9] is not without shortcomings: a 2,000+ image data set is small for CNN; at least 3 filarial species have not been trained in the CNN model, most importantly M. ozzardi; and, finally, a 3D printer is required for the cellphone-microscope adaptor which might hinder roll-out in several areas of the world. This tech is still the future, for example, the computer vision in facial biometrics is present in most cellphones and highly regarded, indicating the future direction of microscopic diagnosis. The authors have kindly provided a guide on how to use it on YouTube [15].