Keywords:
Breast, Oncology, Nuclear medicine, PET-CT, Outcomes analysis, Neoplasia
Authors:
L. Antunovic1, N. Gennaro1, R. De Sanctis1, L. Cozzi1, A. Sagona1, R. Torrisi1, A. Santoro1, R. Zelic2, A. Chiti1; 1Milan/IT, 2Stockholm/SE
DOI:
10.26044/ecr2019/C-0781
Results
Main analysis: In the first model (including clinical data and first-order imaging features),
only age,
molecular subtype,
and HER2 status satisfied the inclusion threshold for the univariable analysis.
Since the molecular subtype is defined based on the expression of selected receptors,
including HER2,
the HER2 status was removed from the model due to possible collinearity issues. Thus,
age and molecular subtype were included in the first model with the actual AUC of 0.71 (table 1).
For the second model (including clinical data,
first-, second- and higher-order imaging features),
one second-order advanced imaging feature (Correlation) and two higher-order imaging features (CoarsenessNGLDM and GLNUGLZLM) satisfied the inclusion threshold for the univariable analysis,
thus were added to the model together with age and molecular subtype.
The actual AUC for this second model was 0.72 (table 2).
Sensitivity analysis: Using the threshold of 0.5 (conservative criterion) for predictor selection,
age,
molecular subtype,
HER2 status,
Correlation,
and CoarsenessNGLDM were selected for the final model.
Age,
molecular subtype,
estrogen receptors,
HER2 status,
type of NAC,
Ki67,
Correlation,
CoarsenessNGLDM,
and GLNUGLZLM were selected when 0.4 was used as the threshold. The actual AUCs for these two models were 0.70 and 0.73,
respectively (table 3).