Skip to main content

Table 4 Association between each identified feature and the tumor size, stratified by detection mode

From: Novel mammographic image features differentiate between interval and screen-detected breast cancer: a case-case study

 

Beta coefficient (p value, 95 % confidence interval), estimated by linear regression modelling

 

Subgroup

Image feature

All cancers (n = 1299)

Screen-detected (n = 927)

Interval cancer (n = 372)

F40 skewness of the intensity gradient

0.16 (p < 0.001, Cl: 0.09 to 0.23)

0.20 (p < 0.001, Cl: 0.12 to 0.29)

-0.05 (p = 0.465, Cl: -0.18 to 0.08)

F21 eccentricity

0.021 (p = 0.56S, Cl: -0.050 to 0.091)

0.055 (p = 0.188, Cl: -0.027 to 0.136)

-0.18 (p = 0.014, Cl: -0.32 to -0.04)

  1. Tumor size was square-root transformed
  2. Feature values calculated based on the dense area of the mammogram as identified by Otsu's method, then Box-Cox transformed and standardized beta coefficients represent the mean change in tumor size (measured in mm) per standard deviation change in transformed feature value Cl - 95 % confidence interval. p values in bold font signifies that they are below 0.05