Validation plot for PLS-DA model (Figure 2) of postoperative samples normalised to the preoperative samples, where 20 permutations have been performed. This plot assesses the risk that the current PLS-DA model is spurious. The goodness of fit () and predictive ability () of the original model are compared with that of several models based on data where the classification parameter (in this case, lAT or hAT) has been randomly permuted, while the original spectral data has been kept intact. The -axis shows the values of and for the original model (on the far right) and the permuted models (further to the left). The -axis shows the correlation between the permuted -vectors and the original -vector for the selected , where represents the parameter used to class the samples. Thus, the original has a correlation of 1 with itself. For the PLS-DA model in this study, the original model has higher and values than the permuted models, and the permuted models all have negative values, indicating that the PLS-DA model is valid.