Use of linear regression and partial least square regression to predict intramuscular fat of pig loin computed tomography images

The intramuscular fat (IMF) content is related to the sensory acceptability of pork, and it can be non-destructively estimated using computed tomography (CT). The aim of this paper is to evaluate the potential use of ordinary linear regression (OLR) of the relative volumes associated with ranges of Hounsfield (HU) values and partial least square (PLS) regression applied to the relative volumes associated with each individual HU value to predict the IMF content using data from one or two different tomograms. The tomograms were obtained from pork loins, and the relative volume associated with each HU value was calculated. The IMF was measured in the loins using a near infrared transmittance device. The best prediction of IMF was obtained by OLR when data from 2 tomograms were used (R2 = 0.83 and RMSEPCV = 0.46%). The results suggest that CT has good potential for measuring the IMF in loins and that the accuracy improves when the data from 2 tomograms were combined. The use of partial volumes as predictors with OLR allows for improved accuracy compared to the use of all of the individual volumes with PLS.

► Intramuscular fat of pork loins can be determined with computed tomography.
► Relative volume associated with HU value ranges as predictors of intramuscular fat
► Higher accuracy with ordinary linear regression than partial least square regression
► Combination of scanning procedures improves accuracy to predict intramuscular fat.