TFinder app: artificial intelligence to diagnose tick fever agents and assess parasitemia/bacteremia in bovine blood smears

Abstract

Due to the importance of diagnosing tick fever (TF) agents and their parasitemia in the field to provide appropriate treatment, the objective of this study was to develop an application capable of detecting the presence or absence of these hemopathogens and calculating parasitemia/bacteremia. Therefore, to create the TFinder app, blood smears from the tip of the tail of cattle naturally infected with TF were prepared to train the artificial neural network (ANN) for Anaplasma marginale and Babesia spp. The ANN was trained with images from different microscope fields and angles. For the validation stage of the ANN, new blood smear images from different animals were inserted. The diagnosis of presence or absence and calculation of parasitemia/bacteremia performed by a human being were compared with that of the ANN. The ANN was trained with 8100 and 2871 images of blood smears containing parasitized erythrocytes with A. marginale and Babesia spp. respectively. In the validation stage, it was used a total of new 1000 for A. marginale and 750 for Babesia spp. There was a strong positive correlation between the A. marginale bacteremia (r = 0.9484; R² = 0.8996; p < 0.0001) values obtained by the human and TFinder app, the same occurred for Babesia spp. parasitemia (r = 0.9650; R² = 0.9314; p < 0.0001). The sensitivity, specificity, and accuracy of TFinder app for the presence or absence of A. marginale were 97.7 %, 86.5 %, and 95.4 %, respectively. While for A. marginale bacteremia it was 89.6 %, 98.7 %, and 98.1 %, respectively. For Babesia spp., the sensitivity, specificity, and accuracy of presence and absence were 98.0 %, 91.1 %, and 96.9 %, respectively. The calculation of Babesia spp. parasitemia was 84.3 %, 99.7 % and 98.9 %, respectively. TFinder app can be used to diagnose and calculate the parasitemia/bacteremia of TF agents in bovine blood smear images.