Scoring Enzootic Pneumonia-like Lesions in Slaughtered Pigs: Traditional vs. Artificial-Intelligence-Based Methods.

Area of Science:

• Veterinary medicine
• Biomedical imaging
• Artificial intelligence

Background:

• Artificial intelligence (AI) methods, particularly convolutional neural networks (CNNs), are increasingly utilized in biomedical sciences, especially for diagnostic imaging.
• CNNs have been recently developed for scoring pleurisy and pneumonia in pig lungs.

Purpose of the Study:

• To evaluate the performance of a CNN in diagnosing pleurisy and pneumonia in pig lungs.
• To compare CNN diagnostic accuracy against the established gold standard of expert veterinary assessment.

Main Methods:

• 441 pig lungs (180 healthy, 261 diseased) were assessed using traditional methods (Madec's and Christensen's grids) as the gold standard.
• Photographs of the lungs were subsequently analyzed by a trained CNN.
• Comparison of CNN scores with veterinarian scores using Spearman's correlation.

Main Results:

• The CNN demonstrated high specificity (95.55%) and good sensitivity (85.05%) in disease detection.
• A strong positive correlation was observed between CNN scores and veterinarian scores (Spearman's coefficient = 0.831, p < 0.01).

Conclusions:

• CNNs show significant potential for accurate and efficient disease scoring in slaughterhouse environments.
• The findings support further research into AI applications for meat inspection and animal disease surveillance.