Relationships between computer-extracted mammographic texture pattern features and BRCA1/2mutation status: a cross-sectional study

Table 3 Ability of trained classifier to distinguish between BRCA1/2 mutation carriers and non-carriers in testing dataset

Training dataset*	Testing dataset		Testing dataset results
Description	Number of non-carriers	Number of carriers	Odds ratio	95% CI	P-value	Odds ratio	95% CI	P-value	OR	95% CI	P-value	AUC	SE
			*Unadjusted*			*Adjusted for age*			*Adjusted for age and PMD*
Percent mammographic density (PMD) alone	30	30	1.02²	(0.99, 1.06)	0.21	1.00²	(0.96, 1.04)	0.96	N/A			0.59	0.07
Features alone¹	30	30	2.00 ³	(1.59, 2.51)	0.02	1.93 ³	(1.53, 2.42)	0.03	1.93 ³	(1.54, 2.43)	0.03	0.68	0.07
Features¹ + PMD	30	30	2.10 ³	(1.67, 2.65)	0.01	2.03 ³	(1.62, 2.56)	0.03	N/A			0.72	0.07

*Training dataset includes 70 non-carriers and 107 BRCA1/2 mutation carriers. ¹Four features were selected by the trained classifier: MinCDF, Energy, AVE, and MaxF (COOC); percent mammographic density was not selected by the trained classifier but was forced into the models where noted. ²Odds ratios, per unit increase in percent mammographic density. ³Odds ratios, per one SD increase in probability score from trained classifier; SD from both models = 0.342. AUC, area under the curve; N/A, not applicable; PMD, percent mammographic density; SE, standard error. P-values <0.05 are shown in bold font.

ISSN: 1465-542X