SAAM-VetNet: an attention-based multi-task framework for animal disease detection and severity grading

Early and accurate detection of animal diseases is critical in veterinary medicine and preclinical research, where timely intervention can influence both animal welfare and experimental outcomes. In this study, we introduce SAAM-VetNet, a novel Severity-Aware Attention-Based Multi-Task deep learning framework designed to simultaneously detect animal diseases and grade their severity from medical images. The proposed architecture integrates a convolutional block attention module to enhance feature localization and contextual representation, coupled with a multi-branch learning strategy for disease classification and severity assessment. We evaluate SAAM-VetNet using two publicly available datasets: the Animal Disease Classification dataset and the Mastitis Disease Detection dataset. Our model achieves superior performance with an accuracy of 91.2% and an F1 score of 89.8%, outperforming established baselines including ResNet18, MobileNetV2, EfficientNet-B0, DenseNet121, and vision transformer (ViT). The results demonstrate that incorporating attention mechanisms and severity-aware multi-task learning significantly enhances model interpretability and diagnostic accuracy, offering a robust tool for automated preclinical model selection and veterinary diagnostics.