Deep learning enables rabies virus detection and antibody quantification in cell culture

Rabies remains a major neglected disease, causing tens of thousands of deaths annually in endemic regions. Cell culture-based assays are central to rabies diagnostics and antibody quantification, yet interpretation relies on manual fluorescence image reading, which is labour-intensive, subjective, and difficult to scale. Here, we present, what is to our knowledge, the first AI-based workflow for detecting rabies virus infection in BHK-21 cell cultures and quantifying antibody titres. Three state-of-the-art architectures-ResNet, Swin Transformer V2, and MambaOut-were trained on a large dataset of 2344 fluorescence images (36,822 annotated patches) capturing biological and technical variability. Evaluation at patch, image, and well levels showed MambaOut consistently outperformed other models, achieving 0.996 accuracy, 0.993 F1-score, and 0.999 AUC-ROC at the well level. To assess applicability, models were tested on unseen images from the WHO/WOAH-recommended fluorescent antibody virus neutralisation assay. MambaOut demonstrated complete agreement with ground truth in predicting antibody concentrations, confirming workflow reliability for vaccine assessment and pet travel schemes. These results establish a foundation for AI-assisted rabies diagnostics and open opportunities for applying deep learning to other cell culture-based viral assays. AI-driven automation for high-throughput image analysis offers a scalable and time-efficient solution for diagnostic laboratories, particularly in endemic countries.