Robustness of quantitative hypoxia PET image analysis for predicting local tumor control

Figures & data

Figure 1. TBR_max in patient groups with local tumor control (N = 15) and local failure (N = 7). The horizontal line depicts the median value.

Table I. Parameters of diagnostic accuracy for different offsets: False positive and true positive rate (FPR and TPR). Positive and negative predictive value (PPV and NPV). Log rank p.

ΔTBRmax	−20%	−10%	0%	+ 10%	+ 20%
FPR	0.13	0.20	0.40	0.53	0.67
TPR	0.14	0.43	0.71	1.00	1.00
PPV	0.33	0.50	0.45	0.47	0.41
NPV	0.68	0.75	0.82	1.00	1.00
p	0.64	0.19	0.14	0.03	0.10

Figure 2. ROC curve for the classification of tumors into local failure/local control groups with varying TBR_max offsets increasing from left to right and a constant TBR_max classification threshold of 2.0. Red crosses depict the estimated accuracies for offsets of -20%, -10%, 0%, + 10% and + 20% (from left to right) compared to reference conditions. The AUC is 0.77.

Figure 3. Examples of possible different background ROI definitions in neck muscles. The mean activities are 6.26 kBq/ml (SD 0.71) and 6.88 kBq/ml (SD 0.62) in the white and black region, respectively. The volumes of the black and white ROI are 3.3 cm³ and 3.9 cm³, respectively.