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

Area of Science:

• Veterinary Medicine
• Artificial Intelligence
• Medical Imaging

Background:

• Early detection of animal diseases is crucial for welfare and research.
• Current diagnostic methods can be time-consuming and require specialized expertise.

Purpose of the Study:

• To introduce SAAM-VetNet, a novel deep learning framework for simultaneous animal disease detection and severity grading.
• To enhance diagnostic accuracy and interpretability in veterinary and preclinical settings.

Main Methods:

• Developed a Severity-Aware Attention-Based Multi-Task (SAAM-VetNet) deep learning framework.
• Integrated a convolutional block attention module and a multi-branch learning strategy.
• Evaluated the model on two public datasets: Animal Disease Classification and Mastitis Disease Detection.

Main Results:

• SAAM-VetNet achieved 91.2% accuracy and an 89.8% F1 score.
• Outperformed established models like ResNet18, MobileNetV2, EfficientNet-B0, DenseNet121, and Vision Transformer (ViT).
• Demonstrated significant enhancement in model interpretability and diagnostic accuracy.

Conclusions:

• Attention mechanisms and severity-aware multi-task learning improve diagnostic performance.
• SAAM-VetNet offers a robust tool for automated veterinary diagnostics and preclinical model selection.