Table 2 Mammograms enhanced the preoperative prediction of lymph node metastasis across all datasets

From: Deep learning on routine full-breast mammograms enhances lymph node metastasis prediction in early breast cancer

Set

Model modality

ROC AUC ± SD

Improvement CI (95%)

p value

Development set (site 1, N = 688, Double CV)

PreopClinic

0.616 ± 0.023

–

–

PreopClinic + FullMammo

0.639 ± 0.023

−0.006–0.050

0.119

PreopClinic + ROIMammo

0.668 ± 0.022

0.018–0.083

0.002

PreopClinic + Tsize&Multifoc

0.718 ± 0.021

0.063–0.140

<0.001

Development set (site 2, N = 259, Double CV)

PreopClinic

0.687 ± 0.039

–

–

PreopClinic + FullMammo

0.748 ± 0.035

0.006–0.108

0.021

PreopClinic + ROIMammo

0.748 ± 0.035

−0.002–0.119

0.071

PreopClinic + Tsize&Multifoc

0.747 ± 0.032

−0.005–0.119

0.077

Independent test (site 2, N = 109)

PreopClinic

0.690 ± 0.063

–

–

PreopClinic + FullMammo

0.774 ± 0.057

0.001–0.154

0.037

PreopClinic + ROIMammo

0.757 ± 0.060

−0.021–0.143

0.142

PreopClinic + Tsize&Multifoc

0.747 ± 0.054

−0.058–0.159

0.300

  1. The mean and standard deviation (SD) of the receiver operating characteristic (ROC) area under the curve (AUC), as well as the confidence interval (CI) for the improvement, were calculated using bootstrap with 1000 samples. All CIs for model improvement and permutation test p values were calculated in comparisons to PreopClinic.
  2. CV cross-validation, N number of patients, PreopClinic preoperative clinicopathology, fullMammo full-breast mammogram, ROIMammo mammogram of region of interest, Tsize tumor size, Multifoc multifocality.
Back to article page