Development and Evaluation of an Open-Source Software Package “CGITA” for Quantifying Tumor Heterogeneity with Molecular Images
Table 3
Comparison of AUC, specificity, and sensitivity of heterogeneity indices vs. SUVmean and TLG.
Parent
Feature
AUC
Sensitivity (%)
Specificity (%)
value†
Intensity-size-zone
Low-intensity short-zone emphasis
0.90*
77.8
88.9
0.004
Intensity-size-zone
Short-zone emphasis
0.81*
77.8
66.7
0.024
Texture Feature Coding Cooccurrence
Contrast
0.72
55.6
88.9
0.085
Intensity-size-zone
High-intensity zone emphasis
0.70
66.7
77.8
0.145
Intensity-size-zone
Zone percentage
0.70
55.6
88.9
0.122
SUV statistics
Entropy
0.70
66.7
77.8
0.145
SUV statistics
Mean SUV
0.60
66.7
66.7
0.453
SUV statistics
Maximum SUV
0.57
66.7
66.7
0.627
SUV statistics
TLG
0.52
55.6
66.7
0.895
denotes that the value is less than 0.05 given the null hypothesis of AUC <0.5. calculated using the Kruskal-Wallis test (19 indices have a value greater than 0.453).