Artificial Intelligence and Citizen Science as a Tool for Global Mosquito Surveillance: Madagascar Case Study

Anopheles stephensi is an invasive and deadly malaria vector with the ability to use artificial containers as larval habitats. This ability is unique for malaria vectors in Africa and requires distinct surveillance strategies for early detection and rapid response. In this study, we trained a variety of artificial intelligence (AI) image recognition algorithms, using thousands of smartphone photos of laboratory-authenticated An. stephensi and seven endemic mosquito species, to develop a citizen science-friendly tool for An. stephensi detection. In Antananarivo, Madagascar, citizen science observations of >132 Anopheles spp. larvae from multiple artificial containers—including one closeup photo of a larva, from a tire—were submitted via NASA’s GLOBE Observer app in March 2020 and discovered years later. Given that genetic testing was no longer possible, this photo was used as a proof-of-concept to determine whether the AI species identification could be used on citizen science-generated images. The tire larva was classified as An. stephensi by all 11 species models, which yielded high accuracy and confidence (up to 99.34%) and included a false positive rate of <1%. Furthermore, explainable AI (XAI) heat maps led to the discovery of dark spots in abdominal segment VI corresponding to testes, corroborating a separate classification of the tire larva as male by the sex model. All available evidence suggests that AI image identification would have flagged this larva as a suspect An. stephensi, which could have been submitted to a molecular laboratory for further confirmation. Results demonstrate the power of integrating citizen science and AI—for which we provide free online tools—as a low-cost signal for malaria programs to confirm and respond to, and as complementary surveillance to fill the critical knowledge gaps in the distribution of invasive An. stephensi across Africa and beyond.