Identifying disease vector images in the Americas in the age of artificial intelligence

TL;DR

This paper explores how AI can help identify disease-carrying vectors in the Americas, focusing on recent advances and challenges in automated identification systems.

Contribution

The study provides a critical analysis of AI-based vector identification systems in the Americas over the last decade.

Findings

Most AI studies focus on mosquito species identification, with limited diversity in vector databases.

Approximately 30 algorithms are used, with AlexNet, MobileNet, and ResNet being the most common.

Only one study in the Americas has developed an AI-based app or device for vector identification.

Abstract

Digital technologies and artificial intelligence (AI) have become integral in many fields, including medicine. Neglected tropical diseases transmitted by vectors, such as arboviral diseases, spotted fever, Chagas disease, and leishmaniasis, pose a significant impact on public health, particularly in the Americas. Strengthening surveillance and control requires the use of digital technology to identify vectors. In this study, we explored how AI can aid in identifying vectors in the Americas and strengthen disease surveillance and control efforts. We reviewed the literature on the automated identification of triatomines, mosquitoes, sand flies, and ticks, focusing on advances in the Americas over the last 10 years, and provided a critical analysis of the automated identification systems for each group. Moreover, we analyzed the development stages of each study: image acquisition, image processing, algorithm training, algorithm testing, app development, app availability, and AI-based devices for vector identification and surveillance. Most studies have applied AI to identify mosquito species. The vector species databases were not diverse, and the most representative group was Triatominae, comprising 65 species (41% of all described species). Currently, approximately 30 algorithms are used for automated vector identification, with the most common being AlexNet, MobileNet, and ResNet. Most studies are in the algorithm training stage, and in the Americas, only one study has progressed to the development of applications or devices. These results highlight the potential of AI for identifying vectors in the Americas, supporting the use of automated visual identification systems as a promising approach to improve vector surveillance, while also promoting citizen science.

Figures