Using latent class modelling to evaluate the performance of a computer vision system for pig carcass contamination

Area of Science:

• Food Science
• Computer Vision
• Statistical Modeling

Background:

• Assessing pig carcass contamination is crucial for food safety.
• Traditional inspection methods have limitations in accuracy and consistency.
• Computer vision systems (CVS) offer automated inspection capabilities.

Purpose of the Study:

• To evaluate the performance of a computer vision system (CVS) for measuring pig carcass contamination.
• To compare the CVS performance against official auxiliary inspections.
• To utilize latent class modeling for performance evaluation without a gold standard.

Main Methods:

• Latent class modeling was employed to estimate sensitivity and specificity.
• Data from 69,215 pig carcasses over 16 days were analyzed.
• Comparison of CVS results with official auxiliary assessments using Cohen's kappa and PABAK.

Main Results:

• The CVS demonstrated a median sensitivity of 31.6% and specificity of 97.9%.
• Official auxiliaries showed a sensitivity of 22% and specificity of 99%.
• Oil contamination was frequently misidentified as faecal contamination by the CVS.

Conclusions:

• The computer vision system excels at identifying true contaminations (high specificity).
• Latent class modeling provides a robust framework for evaluating inspection systems.
• CVS technology can complement traditional meat inspection to improve food safety.