Peeling outcomes of processing tomatoes were predicted using multivariate analysis of magnetic resonance (MR) images. Tomatoes were obtained from a whole-peel production line. Each fruit was imaged using a 7-tesla MR system, and a multivariate data set was created from 28 different images. After imaging, the fruits were individually tagged and processed in a pilot peeling system. An expert grader then assessed the peeling outcome for each fruit; outcomes included “whole peel,”“some skin attached” and seven others. The multivariate analysis techniques of partial least squares-discriminant analysis (PLS-DA) and soft independent modeling of class analogy (SIMCA) were used to predict peeling outcome from the 28 MR images. The PLS-DA model for the “whole peel” (best) outcome correctly classified 81% of the fruit that were in this category. The SIMCA model performed well for rejecting non-“whole peel” fruit but did not perform as well for identification of “whole peel” fruit.


Fruit processors seek methods for predicting the quality of their finished products from measurements of their raw materials (individual fruit). Such predictions will ensure optimal allocation and processing of each fruit, which will result in raw material cost savings, improved product quality, wastewater reduction and energy efficiency. In the case of processing tomatoes, peeling outcome is the metric of interest. This study demonstrates the feasibility of predicting final product quality using MR imaging of raw food materials.