Calibration models for lamb carcass composition analysis using Computerized Tomography (CT) imaging

Dissection of carcasses is a costly, laborious and time-consuming method of assessing carcass tissue composition, and is often inaccurate due to human measurement errors (i.e. cutting error). The need for accurate, non-invasive and objective measurements, both scientifically and industrially, have introduced CT (Computerized Tomography) as a robust, cheaper and less time-intensive tool. Digital images from CT can be used to model carcass tissue composition, introducing direct estimation (Otsu thresholding and Parallel Factor Analysis (PARAFAC)) and multivariate calibration methods (Partial Least Square Regression (PLS).and multi-way PLS (NPLS)). 15 anatomical sites on 120 lamb Norwegian carcasses were CT scanned before they were commercially dissected. The data was separated into calibration (n = 84) and test set (n = 36). The results showed that multivariate calibration using NPLS gave the best results for fat and muscle tissue with respect to prediction error (RMSEP).