Automated identification of spotted-fever tick vectors using convolutional neural networks

Ticks are key ectoparasites for the One Health approach, as they are vectors of pathogens that infect humans, domestic and wild animals. The bacteria Rickettsia rickettsii and R. parkeri are the aetiological agents of tick-borne spotted fever (SF) in South America, where Amblyomma sculptum, A. aureolatum, A. ovale and A. triste are the main vectors. Studies in the medical and biological fields show that artificial intelligence, through machine learning, has great potential to assist researchers and health professionals in image identification practices. The aim of this study was to evaluate the performance of the Convolutional Neural Networks (CNN) AlexNet, ResNet-50 and MobileNetV2 for identifying tick species transmitting SF bioagents. We organised an image database with the following groups: females (368), males (458), dorsal (423), ventral (403), low resolution (328), high resolution (498) and all together (sex+position+resolution = 826), to identify the three main vectors of SF bioagents (Amblyomma aureolatum, A. ovale and A. sculptum), two other possible vectors (A. triste and A. dubitatum) and the species A. cajennense sensu stricto (s.s.), which has similar morphology to A. sculptum but no known vectorial capacity. To evaluate the network's performance, we measured accuracy, sensitivity and specificity. We used Grad-CAM to highlight the regions of the images most relevant to the predictions. CNNs achieved accuracy rates of ~90% in identifying ticks and showed sensitivities of 59%-100% according to species, sex, position or image resolution. When considering all images, both AlexNet and MobileNetV2 recorded the best sensitivity and specificity values in identifying SF vectors. The most relevant areas for classifying species varied according to algorithms. Our results support the idea of using CNNs for the automated identification of tick species transmitting SF bioagents in South America. Our database could support the development of tick identification apps to aid public health surveillance and contribute to citizen science.