Nema et al1 discuss pertinent aspects of artificial intelligence (AI) for malaria. While these algorithms appear ready for deployment given the high median accuracy of 96.7%2 (see Appendix, bubble-plot), malaria film digitalization with present day high-cost digital microscopes poses a significant barrier for adoption, particularly in resource-poor areas. The authors propose using camera smartphones to develop algorithms; however, this has limitations in image standardization and lacks the scalability in digitalizing large numbers of films. To fully harness AI, the development of low cost, automated digital microscopes allowing rapid whole slide imaging is necessary.
Secondly, existing regional malaria slide banks have yet to establish online digital archives. Hence, most AI studies used data exclusively from National Institutes of Health's malaria database (27 558 images, falciparum and vivax species only), with few trained from private datasets. These studies, including a large multicentre study3 evaluating an algorithm trained with 500 slides from 11 countries, lack high quality data of uncommon, but morphologically distinct species of ovale, malariae and knowlesi. In a post COVID-19 world with disruption of existing malaria control measures4 and increasing cross-border travel, acquired non-falciparum, non-vivax malaria infections may be missed by AI due to biased datasets.5 We propose existing regional malaria slide banks digitize their slides to provide publicly available digital datasets to augment data of uncommon malaria species. Lastly, the establishment of a global digital repository can provide oversight in image standardization for training, downstream validation, and quality assurance for robust AI solutions for malaria.
Contributors
All authors contributed equally to the data analysis and writing of this correspondence.
Declaration of interests
The authors declare no competing interest. B.E. Fan is supported by the National Medical Research Council (NMRC) Clinician Innovator Development Award (NMRC/CIDA19May-0004) for his work on Algorithm for diagnosis of peripheral blood films via deep learning and computer vision.
References
- 1.Nema S, Rahi M, Sharma A, Bharti PK. Strengthening malaria microscopy using artificial intelligence-based approaches in India. Lancet. 2022;5 doi: 10.1016/j.lansea.2022.100054. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Fan BE, Wang SSY, Natalie Aw MY, et al. Artificial intelligence in peripheral blood films: an evolving landscape. Lancet Haematol. 2022;9(3):e174. doi: 10.1016/S2352-3026(22)00029-1. [DOI] [PubMed] [Google Scholar]
- 3.Horning M.P., Delahunt C.B., Bachman C.M., et al. Performance of a fully-automated system on a WHO malaria microscopy evaluation slide set. Malar J. 2021;20:110. doi: 10.1186/s12936-021-03631-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Zawawi A, Alghanmi M, Alsaady I, Gattan H, Zakai H, Couper K. The impact of COVID-19 pandemic on malaria elimination. Parasite Epidemiol Control. 2020;11:e00187. doi: 10.1016/j.parepi.2020.e00187. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Ricci Lara M.A., Echeveste R., Ferrante E. Addressing fairness in artificial intelligence for medical imaging. Nat Commun. 2022;13:4581. doi: 10.1038/s41467-022-32186-3. [DOI] [PMC free article] [PubMed] [Google Scholar]