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

Table 2 Descriptive characteristics of selected computer-extracted features

			Correlation with percent mammographic density (n = 237)		Correlation with age (n = 237)		Non-carriers (n = 100)		UnaffectedBRCA1/2carriers (n = 137)
Feature type and number	Feature¹	Definition	r*	P-value	r*	P-value	Mean (SD)	Range	Mean (SD)	Range
Gray-level magnitude-based features:
M1	AVE	Average gray value within ROI; higher values correspond to denser region	0.31	<0.0001	0.01	0.89	139.6 (26.1)	69.0, 223.5	134.1 (28.9)	59.0, 242.0
M2	MinCDF	Gray value corresponding to the 5% region cutoff on cumulative density function; higher values correspond to denser region	-0.13	0.04	0.23	0.0005	98.0 (23.6)	35.0, 162.0	78.1 (27.3)	15.0, 210.0
M3	Balance	Ratio of (95% CDF-AVE) to (AVE-5% CDF); related to skewness; values less than one correspond to having an ROI that is skewed toward relatively denser values	-0.32	<0.0001	-0.04	0.49	1.07 (0.42)	0.45, 3.31	1.11 (0.40)	0.38, 2.33
Texture-based features:
T1	Energy	Measure of image homogeneity; higher values correspond to being more homogeneous	-0.30	<0.0001	0.19	0.003	0.004 (0.011)	0.0, 0.109	0.003 (0.004)	0.0, 0.028
T2	MaxF (COOC)	Largest number of a gray value pair in the co-occurrence matrix; measure of image homogeneity; higher values correspond to being more homogeneous	-0.24	0.0002	0.15	0.02	0.012 (0.024)	0.001, 0.239	0.010 (0.017)	0.001, 0.145

CDF, cumulative density function; COOC, co-occurrence; ROI, region-of-interest. *Spearman’s rank correlation coefficient. ¹All features were selected using the training dataset. The Balance feature was only selected in sensitivity analyses where the training dataset was truncated at the upper age-limit of mutation carriers. P-values <0.05 are shown in bold font.

ISSN: 1465-542X