Fig 1.
Summary of the workflow implemented for the estimation of the heterogeneity in the PET images: 1) The Volume of Interest (VOI) corresponding to the tumor is extracted and the image quantized to 64 levels; (a) Quantitative metrics are measured: 2a. The metabolic parameters described in Table 1; 3a. The texture features–First order, local (Gray Level Co-occurrence Matrix (GLCM)), regional (Gray Level Run Length Matrix (GLRLM)).
Table 1.
List of computed features from FDG-PET.
Fig 2.
Major axial plane of the extracted VOI from four of the tumors analyzed with VOI boundaries shown in yellow. (a) and (c) show an example of homogeneous tumors with zero score in the visual heterogeneity scale; (b) and (d) show an example of heterogeneous tumors with score one in the visual heterogeneity scale. (e) and (g) show an example of tumors with zero score in the visual pattern scale; (f) and (h) show an example of tumors with score one in the visual pattern scale.
Fig 3.
Comparison of the most representative slide for each patient from the resampled scans used for the heterogeneity analysis.
Each slide was selected to contain the SUVmax in the tumour VOI of each patient. Non-responders and responders are located in the left and right sides while the vertical order is given by decreasing uptake–selected to be descending SUVmean.
Table 2.
Comparison of the baseline clinical and immunohistochemistry (IHC) characteristics of the patients.
The p-value corresponds to the χ2 test for gender and clinical staging risk group (degrees of freedom are 35 in both cases) and to the Mann-Whitney U test in the rest of variables.
Table 3.
Comparison of the visual scoring system and PET parameters in responders and non-responders together with the χ2 test results and Mann-Whitney U test results respectively.
Fig 4.
Comparison of the ROC curves using the different sets of features proposed to predict tumor response.