Table 1.
Summary of the studied pathologies.
Fig 1.
Three representative benign and malignant spectra.
On visual inspection no particular peaks or patterns can be identified that predict benign or malignant character of the lymph nodes. In particular, lactate, choline and creatine peaks, which are commonly used in clinical radiology, do not give obvious predictive information.
Fig 2.
Score plot of training set differentiating benign and malignant causes of lymphadenopathy with cross-validation plot.
A. t1 is the predictive component, it is used to achieve discrimination in between both groups. t1o is the orthogonal component, It is useful for understanding class variability. B. Cross-validation plot showing Tcv1 vs Tcv2 for the OPLS-DA model with three orthogonal components showing good separation of the benign and malignant groups (R2 = 0.96, Q2Y = 0.63, n = 45).
Fig 3.
Blinded test set loaded in model predicting benign and malignant tissues.
New benign and malignant samples not previously analyzed were blindly loaded into the OPLS-DA model from Fig 2, and the categories were subsequently revealed. t1 and t1o are the same as in Fig 2.
Fig 4.
Correlation between ppm regions and the predictive component.
The average of all of the studied spectra is represented along the ppm axis. This graph allows visualization of the correlation between each ppm region and the predictive t1 score. The orientation of the peak with regards to the axis indicates if they are benign predictors (above the x axis) or malignant predictors (below the x axis). Color coding corresponds to the relative weight of each peak in predicting benign vs. malignant classes. A red color indicates peaks that have a high predictive value while a blue color indicates one that has a lower predictive value. Note the peak with the highest predictive value is found at 3.8ppm.
Fig 5.
Blinded test set loaded in model predicting age of tissue.
Time since collection is expressed in months. There is no separation of the samples which indicates that the age of the sample has no predictable effect of the spectrum.