AI enhances tick risk identification and prediction

The use of artificial intelligence (AI) is driving significant change in the identification and surveillance of ticks and the diseases they transmit. New machine learning tools have the potential to increase diagnostic accuracy and speed, impacting control strategies, risk prediction, and integrated management of vector-borne diseases in humans and animals.

Accurate identification of tick species is a central challenge for the surveillance of diseases transmitted by these vectors. Traditional methods, based on morphology, require trained specialists and can be imprecise when dealing with closely related species. A study led by Sebai and colleagues developed an experimental deep learning prototype to automatically recognize five Hyalomma species, which are important in veterinary and human medicine in Tunisia. Using 1,344 images of morphologically identified ticks, the YOLOv11-CNN system achieved accuracy between 97% and 100% in species classification, but only under controlled laboratory conditions and for a limited set of species. Explainability analysis showed that the model's decisions were based on relevant morphological regions traditionally used by taxonomists, indicating that AI can replicate and even expand the human perspective in taxonomic identification.

Beyond identification, AI models are being applied to predict risk areas and tick abundance, and consequently, diseases such as Lyme disease and tick-borne encephalitis. Kelly and collaborators used Random Forest algorithms to analyze data from 93,289 ticks of 11 species collected in Japan between 1990 and 2023. The model, fed by 26 environmental variables, achieved area under the curve (AUC) above 0.89 in predicting suitable habitats for Ixodes ovatus and Ixodes persulcatus, the main vectors of these diseases in the country. The study identified high suitability in Hokkaido and cold, humid regions of Honshu, suggesting that risk areas may be broader than previously thought. According to the authors, “climatic variables were the strongest predictors, with more than 90% cumulative importance in the model.”

Another important advance is the ability of AI models to integrate environmental and biological factors to predict not only tick abundance but also disease incidence in humans. Angell and colleagues compared traditional statistical methods with machine learning (gradient boosting) to predict tick abundance and Lyme disease cases in Minnesota, United States. The AI models, which incorporated climate and small mammal fauna data, showed higher predictive accuracy one year in advance for tick abundance and two years for Lyme cases. The AUC was higher in models with time lag, suggesting that current environmental factors are good predictors of future risks. The authors highlight that “monitoring environmental factors offers expanded opportunities for public health interventions through prediction of tick abundance and possible consequences for increased Lyme disease incidence.”

The use of AI also extends to the surveillance of bacterial pathogens associated with veterinary and aquatic environments. Amulraj and collaborators presented a model based on surface-enhanced Raman spectroscopy (SERS) and machine learning for virulence profiling of Acinetobacter baumannii, a multidrug-resistant bacterium relevant to animal and human health. The system, tested on 20 environmental and veterinary isolates, achieved realistic accuracy of 87.9% in classifying between virulent and avirulent isolates, according to leave-one-strain-out validation. The study highlights that information on geographic origin and host is partially embedded in the spectral signatures, which can enhance integrated surveillance in One Health programs.

Despite advances, there are important limitations. Automated tick identification models, such as that of Sebai and colleagues, were evaluated only under controlled laboratory conditions and with geographically limited samples, which may restrict their applicability in the field. Studies like that of Kelly and collaborators depend on the quality and scope of surveillance data, and transferring models to other regions requires new validations. In bacterial surveillance, Amulraj and team warn of the need for validations aware of strain profiles to avoid overestimating accuracy. In addition, Hirata and colleagues emphasize that although genomic and AI technologies facilitate the detection of new tick-borne viruses, establishing causal relationships between viral presence and disease still requires traditional pathological approaches, such as in situ localization of agents in tissues.

In practice, incorporating AI into surveillance systems can speed up diagnosis, standardize vector identification, and anticipate high-risk areas and periods, enabling faster and more targeted responses in public and veterinary health. Automated tools can support laboratories with few specialists and expand monitoring coverage, while predictive models based on climate and fauna assist in planning preventive campaigns. The integration of traditional and modern approaches, as suggested by Angell and Hirata, is essential to advance the understanding and control of tick-borne diseases within a One Health perspective.

The advancement of artificial intelligence tools for identification, surveillance, and risk prediction related to ticks and vector-borne diseases represents a paradigm shift in public and veterinary health. Although technical and validation challenges persist, the results already obtained point to a future in which AI will be central to integrated control strategies for these diseases, enhancing the response to emerging threats on a global scale.