. Author manuscript; available in PMC: 2022 Oct 1.

Published in final edited form as: Med Image Anal. 2021 Jul 2;73:102138. doi: 10.1016/j.media.2021.102138

Table 2:

Case-control discrimination performance on the dataset ds3-b for breast percent density (PD) values generated by Deep-LIBRA and LIBRA, area-based (A Quantra) and volumetric (V Quantra) PD values by Quantra, and clinical BI-RADS density assessments. Results correspond to mean AUCs and 95% CIs in parentheses. Folds 1, 2 and 3 are the held-out folds used for the evaluation of the corresponding Deep-LIBRA SVM. Unadj. and adj. indicate unadjusted logistic regression models and logistic regression models adjusted for age and BMI, respectively.

	1 LIBRA	V_Quantra	A_Quantra	BI-RADS	Deep_LIBRA
Fold l unadj.	0.469 (0.465, 0.472)	0.578 (0.578, 0.579)	0.560 (0.559, 0.560)	0.541 (0.540, 0.541)	0.532 (0.528, 0.536)
Fold 2 unadj.	0.460 (0.455, 0.465)	0.579 (0.578, 0.579)	0.560 (0.559, 0.560)	0.541 (0.540, 0.541)	0.594 (0.593, 0.594)
Fold 3 unadj.	0.467 (0.463, 0.471)	0.578 (0.578, 0.579)	0.561 (0.559, 0.561)	0.541 (0.541,0.541)	0.561 (0.560,0.561)

All unadj.	0.467 (0.464, 0.471)	0.579 (0.579, 0.580)	0.561 (0.560, 0.561)	0.540 (0.539, 0.542)	0.578 (0.577, 0.578)
All adj.	0.498 (0.494, 0.502)	0.586 (0.584, 0.587)	0.568 (0.567, 0.570)	0.550 (0.548, 0.552)	0.582 (0.581,0.583)