A Systematic Review of Artificial Intelligence Approaches for Foot-and-Mouth Disease Detection in Low-Resource Cattle Systems

Foot and Mouth Dise(FMD) is a highly contagious transboundary livestock disease causing severe economic losses in low- and middle-income countries. In low-resource cattle systems characterized by inadequate veterinary infrastructure, limited laboratory access, and poor connectivity, traditional diagnostic methods are often unavailable or delayed. Artificial Intelligence (AI) approaches have emerged as potential solutions for rapid, accessible, and low-cost FMD detection. However, no systematic review has specifically synthesized evidence on AI approaches for FMD detection in resource-constrained cattle systems. This systematic review aimed to identify, characterize, and evaluate current AI approaches for FMD detection in cattle, with specific focus on methods suitable for low-resource settings; synthesize performance metrics; analyze data quality challenges; and assess deployment readiness. Following PRISMA 2020 guidelines, we systematically searched Scopus, Web of Science, PubMed, and Google Scholar for peer-reviewed studies published between January 2020 and August 2026. Studies were included if they described AI-based FMD detection in cattle, reported quantitative performance metrics, and addressed explicit or implicit resource constraints. Risk of bias was assessed using a modified QUADAS-AI tool. Due to heterogeneity in study designs and outcome measures, results were synthesized narratively. Of 1,247 records screened, 24 studies met inclusion criteria. Three main AI approach categories emerged: (1) computer vision systems (n=14) using convolutional neural networks and YOLO architectures, achieving detection accuracies of 88-98%; (2) structured data predictive models (n=7) using gradient boosting and neural networks on production records, achieving F1-scores of 75-88%; and (3) hybrid and multimodal frameworks (n=3) integrating images, production data, and environmental parameters, achieving accuracies of 85-94%. Key data quality challenges included class imbalance (reported in 79% of studies), label noise (54%), and limited dataset sizes (67%). Mitigation strategies included noise detection algorithms, synthetic oversampling (SMOTE variants), transfer learning, and multimodal fusion. Only two studies reported prospective validation in field conditions, and only one assessed end-user usability. AI approaches for FMD detection in low-resource cattle systems have demonstrated promising performance under controlled conditions. However, the evidence base is limited by retrospective validation, lack of field deployments, and insufficient attention to user-centered design. The integration of multimodal data and agentic AI capabilities represents an emerging frontier with potential to address the unique constraints of low-resource settings. Future research should prioritize prospective validation studies, standardized reporting, explainable AI approaches, and integration with existing veterinary surveillance workflows.