Direct decomposition of NMR relaxation profiles and prediction of sensory attributes of potato samples

In this paper the decomposition of low-field Carr-Purcel-Meiboom-Gill (CPMG) NMR relaxation measurements on 23 raw potato categories was investigated. The potato categories were formed from five different cultivars, each binned in 2 or 3 dry matter intervals, sampled at two storage times. A novel data analytical tool-called SLICING-revealed that different amounts of four distinct proton relaxation profiles could describe the main variation in the data set. Magnitudes (scores) of the third and fourth profile separated the potato cultivars, storage times, and dry matter content indicating that properties related to fast relaxation times explain the differences between cultivars and storage times for the potatoes. The concept of direct decomposition using SLICING on low-resolution NMR data is a new approach in potato analysis and a promising tool for obtaining more information about the structure and water distribution in food products. Furthermore, the texture-related sensory attributes, hardness, cohesiveness, adhesiveness, mealiness, graininess, and moistness of cooked potatoes were predicted by partial least-squares regression (PLSR). Four different types of predictor variables derived from the NMR relaxation curves were compared in the regression models: (i) the raw CPMG curves, (ii) the parameters from the traditional bi-exponential fitting; (iii) the results from a distribution analysis, and (iv) the scores from the SLICING model. The predictions based on the distribution analysis performed worse than the first three procedures, which all showed similar prediction ability. The advantage of the SLICING approach is in the possibility to interpret physical properties, e.g. water distribution of the potato samples. (C) 2003 Swiss Society of Food Science and Technology. Published by Elsevier Science Ltd. All rights reserved.