Successful differentiation of normal chicken livers from septicemic chicken livers was demonstrated using
visible/near-infrared (Vis/NIR) spectral data subjected to principal component analysis and then fed into a feedforward
back-propagation neural network. The study used 300 fresh chicken livers, 150 collected from normal
chicken carcasses and 150 collected from chicken carcasses diagnosed with the septicemica/toxemia (septox)
condition as defined for condemnation under U.S. Department of Agriculture (USDA) standards for food safety.
Using a training set of 200 samples and testing set of 100 samples, the best neural network model demonstrated a
classification accuracy of 98% for normal samples and 94% for septicemia/toxemia samples. These results show
that Vis/NIR spectral methods have potential for use in chicken liver inspection as part of automated online systems
for food safety inspection. Liver abnormalities are identifying characteristics of the septox condition; consequently,
liver screening would be extremely useful as part of an automated inspection system to meet USDA food safety
requirements for poultry. Automated inspection systems capable of real-time on-line operation are currently being
developed, and spectroscopic liver inspection is potential tool that could be implemented as part of such systems to
help poultry processors increase production while meeting food safety inspection requirements.