Skip to main content
. 2026 Aug 24;131:106442. doi: 10.1016/j.ebiom.2026.106442

Table 2.

ULM features considered significant after Benjamini-Hochberg correction.

Feature p pCR Non-pCR Effect size Accuracy (%) AUC
Measurement 1 → 2
 Compensated vessel coverage in % (|ΔF|) 0.0005 28.39 ± 12.52 7.50 ± 6.34 2.21 [0.48, 3.43] 84 0.95 [0.71, 1.00]
 Velocity range—skewness (Fmin) 0.0014 0.66 ± 0.11 1.14 ± 0.51 −1.13 [−1.67, 0.70] 85 0.94 [0.71, 1.00]
 Tortuosity (SOAM)—mean in rad/mm (Fmin) 0.0018 6.07 ± 0.40 7.93 ± 1.85 −1.21 [−1.99, −0.29] 90 0.93 [0.65, 1.00]
 Track length—median in mm (Fmax) 0.0018 0.57 ± 0.04 0.44 ± 0.09 1.49 [0.81, 2.19] 80 0.93 [0.67, 1.00]
 Distances—skewness (|ΔF|) 0.0018 1.41 ± 0.52 0.58 ± 0.54 1.49 [0.15, 2.88] 68 0.92 [0.62, 1.00]
 Compensated vessel coverage in % (Fmax) 0.0023 41.71 ± 11.45 16.06 ± 14.57 1.83 [0.71, 3.42] 80 0.92 [0.69, 0.99]
 Tortuosity (SOAM)—median in rad/mm (Fmin) 0.0049 3.92 ± 0.56 5.80 ± 2.09 −1.08 [−1.60, −0.33] 75 0.89 [0.60, 0.98]

Given for an FDR of 5% (highlighted in bold) or 10%. From left to right: feature name, nominal p-value, mean ± standard deviation of patients with pCR and non-pCR, effect size (Hedges' g with bootstrapped 95% confidence interval), cross-validated predictive accuracy, AUC with bootstrapped 95% confidence interval.