A New Method to Detect Buffalo Mastitis Using Udder Ultrasonography Based on Deep Learning Network.

Area of Science:

• Veterinary Medicine
• Artificial Intelligence
• Animal Science

Background:

• Mastitis significantly impacts dairy production globally, causing reduced milk yield, quality issues, and increased costs.
• Early detection of mastitis is crucial for minimizing economic losses and preventing widespread outbreaks in buffalo herds.
• Deep learning offers advanced capabilities for accurate disease diagnosis from medical imaging.

Purpose of the Study:

• To develop and validate a deep learning network for detecting mastitis at the quarter level in buffaloes.
• To assess the model's performance using ultrasound images and varying somatic cell count (SCC) thresholds.
• To provide an accessible diagnostic tool for smallholder farmers in developing countries.

Main Methods:

• Utilized 3054 udder ultrasound images from 271 buffaloes, divided into training, validation, and test sets.
• Employed the EfficientNet_b3 model combined with the Convolutional Block Attention Module (CBAM).
• Addressed class imbalance using the PolyLoss function and evaluated performance with SCC thresholds of 2x10^5 and 4x10^5 cells/mL.

Main Results:

• The deep learning model achieved 70.02% accuracy, 77.93% specificity, 63.11% sensitivity, and an AUC of 0.77 with an SCC threshold of 2x10^5 cells/mL.
• Performance improved with an SCC threshold of 4x10^5 cells/mL, yielding 75.93% accuracy, 80.23% specificity, 70.35% sensitivity, and an AUC of 0.83.
• The model demonstrated superior classification for mastitis when defined as SCC ≥ 4x10^5 cells/mL.

Conclusions:

• A deep neural network model was successfully established for quarter-level mastitis detection in buffaloes.
• The developed model, particularly with an SCC threshold of 4x10^5 cells/mL, is suitable for on-farm mastitis diagnosis.
• This study provides a valuable theoretical basis for mastitis diagnostics in buffaloes, especially for resource-limited settings.