18F-FET PET Brain Metastases: Statistical Analysis

Data preparation, descriptive analysis, and diagnostic performance

Author

Lu Mao

Published

August 31, 2026

1 Analysis objectives

The primary objective is to assess how accurately \(^{18}\)F-FET PET differentiates tumor progression (POD) from treatment-related change (TRC) in previously treated brain metastases. The reference diagnosis is the variable Truth. The four prespecified primary PET predictors are dynamic TBRmax, dynamic TBRmean, static TBRmax, and static TBRmean.

The diagnostic outcome is coded with TRC as the reference level and POD as the positive class.

2 Study population and descriptive statistics

2.1 Cohort characteristics

The primary purpose of this table is to describe the analysis cohort. Because many imaging variables are skewed, continuous measures are summarized using median (Q1, Q3).

Selected characteristics of the study cohort
Characteristic Overall
Subjects, N 101
POD, n (%) 28 (27.7%)
TRC, n (%) 73 (72.3%)
Age at FET-PET, median (Q1, Q3) 59.0 (51.0, 66.0)
Dynamic TBRmax, median (Q1, Q3) 2.27 (1.82, 2.80)
Dynamic TBRmean, median (Q1, Q3) 1.90 (1.55, 2.29)
Static TBRmax, median (Q1, Q3) 2.40 (2.06, 2.83)
Static TBRmean, median (Q1, Q3) 1.87 (1.74, 1.98)

2.2 Primary PET measures by POD/TRC status

Primary PET measurements by reference diagnosis
Reference diagnosis N Median (Q1, Q3)
Dynamic TBRmax
TRC 72 2.08 (1.72, 2.50)
POD 28 2.86 (2.25, 3.51)
Dynamic TBRmean
TRC 72 1.82 (1.50, 2.22)
POD 28 2.28 (1.86, 2.88)
Static TBRmax
TRC 73 2.24 (1.93, 2.63)
POD 28 2.96 (2.55, 3.54)
Static TBRmean
TRC 73 1.78 (1.72, 1.93)
POD 28 1.98 (1.89, 2.10)

2.3 Distribution of PET measures by reference diagnosis

These displays are descriptive. The principal inferential target is diagnostic discrimination, quantified by the ROC AUC, rather than a separate hypothesis test of POD-versus-TRC location for each PET measure.

3 Primary diagnostic-performance analysis

For each PET measure, the ROC curve is estimated with POD as the positive condition. AUC confidence intervals use DeLong’s method. The operating threshold is selected by maximizing Youden’s index. Sensitivity and specificity are calculated at that threshold.

Measure N POD / TRC AUC (95% CI) Cutoff Sensitivity (95% CI) Specificity (95% CI)
Dynamic TBRmax 100 28 / 72 0.764 (0.659, 0.870) 2.75 57.1% (37.2, 75.5) 84.7% (74.3, 92.1)
Dynamic TBRmean 100 28 / 72 0.734 (0.623, 0.844) 2.34 50.0% (30.6, 69.4) 86.1% (75.9, 93.1)
Static TBRmax 101 28 / 73 0.819 (0.735, 0.902) 2.35 92.9% (76.5, 99.1) 61.6% (49.5, 72.8)
Static TBRmean 101 28 / 73 0.820 (0.739, 0.901) 1.88 89.3% (71.8, 97.7) 67.1% (55.1, 77.7)