Using automated texture features to determine the probability for masking of a tumor on mammography, but not ultrasound

Table 2 Prediction of PMD

Method	MSE	R ²	N
Univariate selection	117.0 (8.6)	0.67 (0.02)	132.5 (9.7)
Lasso	111.9 (8.4)	0.69 (0.02)	108.8 (12.9)
Boosting	113.0 (8.6)	0.68 (0.02)	126.1 (8.8)
Random forest	120.2 (9.7)	0.66 (0.03)	–^a

Summary statistics (mean and standard deviation) of mean squared error (MSE) and R ² obtained from (linear) regression models with selected features, as well as the number of selected features N, are shown. All measurements were obtained by 3-fold cross-validation with 100 repetitions
MSE mean squared error, PMD percentage mammographic density
^aThere was no variable selection with random forest

ISSN: 2047-783X