AfriTickID: An AI-assisted image-based approach for tick species identification in Kenya.

Tick-borne diseases such as Lyme disease, tick-borne encephalitis and Crimean-Congo haemorrhagic fever in humans, and East Coast Fever and anaplasmosis in animals, pose a significant threat to public and veterinary health. Knowing where tick species are distributed helps with targeted response. In this study, we used Artificial Intelligence (AI) to automate the identification of tick species from images. A Kenya-based tick image dataset comprising of 864 images across seven tick species was compiled from laboratory collections and the Global Biodiversity Information Facility. MobileNetV2 and ResNet-50 were used to identify tick species and cross-validated to determine performance and accuracy. We evaluated model performance using 5-fold cross-validation and summarized the results with a normalized confusion matrix. The lightweight MobileNetV2 achieved a better performance than ResNet-50, with an accuracy of 75%, weighted F1-score of 74%, and Kappa score of 69%. MobileNetV2 had the highest identification rates for Rhipicephalus decoloratus (88%) and Rhipicephalus sanguineus (87%), and the lowest for Rhipicephalus evertsi (22%). ResNet-50 and baseline convolutional neural networks performed better on Rhipicephalus sanguineus (94%), with lower rates across other species. These results provide a baseline for AI-assisted tick identification in Kenya. Future work should focus on expanding the dataset to improve model performance for tick species in Kenya, while broader applicability to other regions, species, life stages, and image sources will require larger, balanced datasets and external validation.