Skip to main content
. Author manuscript; available in PMC: 2021 Mar 15.
Published in final edited form as: Pac Symp Biocomput. 2021;26:285–296.

Table 2.

Ten-Fold AUROC statistics for unpenalized logistic regression prediction model on held out test data across all slides for colon (n=172) and LN (n=84); three head columns indicate whether advanced staging was predicted using aggregates of colon sub-compartment assignments (Region Counts); invasion (Tumor Invasion Scores); or Both to lend complementary information; models with main effects and interactions were considered; confidence assessed via 1000-sample non-parametric bootstrap

AUC (10-Fold CV) Region Counts Only Tumor Invasion Scores Only Both
Stage > 2; LN Positive Main Effects Second Order Terms Main Effects Second Order Terms Main Effects Second Order Terms
Colon GNN Pretrained 0.89±0.028 0.85±0.031 0.84±0.032 0.89±0.028 0.91±0.022 0.85±0.031
Colon GNN Not Pretrained 0.89±0.027 0.88±0.028 0.86±0.029 0.89±0.027 0.91±0.023 0.88±0.028
LN GNN Pretrained 0.76±0.077 0.88±0.046 0.88±0.046 0.76±0.077 0.88±0.046 0.88±0.046
LN GNN Not Pretrained 0.89±0.051 0.92±0.039 0.92±0.034 0.89±0.051 0.92±0.037 0.92±0.039