Table 2: Performance of the ConvNet HUP and ConvNetU HCMC/CWRU classifiers on the CINJ data cohort in terms of means and standard deviation of Dice coefficient, PPV and NPV.

From: Accurate and reproducible invasive breast cancer detection in whole-slide images: A Deep Learning approach for quantifying tumor extent

GroupNDicePPVNPV
ConvNetHUP
 All cases400.6771 ± 0.24450.6464 ± 0.28700.9709 ± 0.0350
 Only invasive190.7578 ± 0.21660.7462 ± 0.24800.9654 ± 0.0355
 Mixture210.6041 ± 0.25010.5560 ± 0.29530.5560 ± 0.2953
ConvNetUHCMC/CWRU
 All cases400.6596 ± 0.25270.6370 ± 0.29410.9663 ± 0.0421
 Only invasive190.7596 ± 0.20740.7499 ± 0.24230.9614 ± 0.0440
 Mixture210.5691 ± 0.26020.5348 ± 0.30450.9708 ± 0.0409
  1. The results in Table 2 are organized in terms of all cases in the CINJ cohort (N = 40), a subset of the CINJ cohort with invasive breast cancer alone (N = 19), and a mixture of invasive and other in situ subtypes of breast cancer (N = 21).