Deep learning and high image processing for diagnosis of Zoonotic Bovine Tuberculosis (Mycobacterium Bovis) in Cattle

Abstract

Background: Zoonotic tuberculosis (zTB), caused by Mycobacterium bovis, remains a persistent public health and economic burden in low- and middle-income countries (LMICs). Conventional diagnostic approaches such as tuberculin skin testing and post-mortem inspection are slow, costly, and often unreliable in rural abattoirs. These limitations lead to underreporting and delayed interventions, increasing the risk of human infection through contaminated meat and milk. Recent advances in deep learning, particularly Convolutional Neural Networks (CNNs), offer opportunities to improve diagnostic speed and accuracy in resource-constrained settings. Aim: This study developed and evaluated a CNN-based diagnostic tool for detecting zTB from high-resolution lymph node images of cattle, designed to enhance abattoir-level and veterinary diagnosis in Uganda. Methods: A dataset of 278 lymph node images (zTB-positive and negative) was compiled from the International Livestock Research Institute (ILRI), European Reference Laboratory, and Makerere University’s College of Veterinary Medicine. Images were preprocessed—resized, normalized, and augmented—to ensure consistency and improve robustness. An EfficientNet-B0 CNN, pre-trained on ImageNet, was fine-tuned and evaluated using 70% training, 15% validation, and 15% testing splits. Model performance was measured using accuracy, precision, recall, F1-score, specificity, and AUC-ROC. A web-based interface was developed to demonstrate real-time diagnostic capability. specificity, F1-score, and AUC-ROC. A web-based interface was also developed to demonstrate the model’s potential for real-time diagnostic application. Results: The model achieved over 90% accuracy, sensitivity, and specificity on the test set, with an AUC-ROC above 0.95, confirming excellent discriminative ability. The web interface allowed users to upload images and obtain predictions instantly, demonstrating field-level feasibility. Conclusion: The CNN-based tool significantly improved the accuracy and efficiency of zTB diagnosis compared to conventional methods. By enabling real-time abattoir-level screening, this innovation enhances food safety, strengthens zoonotic surveillance, and supports the One Health approach in Uganda. Keywords: Zoonotic tuberculosis, Mycobacterium bovis, Convolutional Neural Networks, Deep Learning, Veterinary Diagnostics, Image Processing