Image Classification and Automated Machine Learning to Classify Lung Pathologies in Deceased Feedlot Cattle

Area of Science:

• Veterinary Medicine
• Animal Pathology
• Machine Learning Applications

Background:

• Bovine respiratory disease (BRD) and acute interstitial pneumonia (AIP) are major causes of morbidity and mortality in feedlot cattle.
• Bronchopneumonia with an interstitial pattern (BIP) is an emerging feedlot lung disease requiring accurate diagnosis.
• Necropsies are crucial for diagnosing lung diseases and improving feedlot management but face logistical challenges.

Purpose of the Study:

• To develop and evaluate machine learning models for diagnosing respiratory syndromes in feedlot cattle lungs using necropsy images.
• To assess the diagnostic accuracy of image classification models based on gross and histopathological diagnoses.
• To explore the utility of image analysis in overcoming challenges associated with traditional necropsies.

Main Methods:

• Four datasets were created from unaltered and cropped lung images, labeled with either gross or histopathological diagnoses.
• Image classification models were developed for each dataset.
• Model performance was evaluated based on accuracy, with the best trial selected for each model.

Main Results:

• Diagnostic accuracies for gross diagnoses ranged from 39-41% for both unaltered and cropped images.
• Labeling with histopathology diagnoses did not significantly improve average accuracies (34-38%).
• Sensitivities were higher for BIP (60-100%) and BRD (20-69%) compared to AIP (0-23%).

Conclusions:

• The developed machine learning models require further fine-tuning for improved diagnostic accuracy.
• These models represent a preliminary step towards aiding veterinarians in diagnosing cattle lung diseases during field necropsies.
• Image-based diagnostics offer a potential solution to logistical constraints in traditional necropsy procedures.