Skip to main content

Table 5 Discovery rates for three models and different cut-off points

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

Cut-off point for predicted masking risk (%)a Frequency above cut-off point (%)b Discovery rates for tumors not seen on mammography (%)
Clinical modelc Boosting PMD modeld Observed PMD modele
5 47.5 81.8 80.9 78.9
10 26.4 54.5 57.7 55.6
12 20.0 44.8 47.4 48.7
15 13.6 32.9 35.1 39.7
20 7.5 16.2 21.0 25.4
  1. All measurements were obtained by 3-fold cross-validation with 100 repetitions
  2. BMI body mass index, HRT hormone replacement therapy, PMD percentage mammographic density
  3. aPatients were classified into a “high-risk” group if the prediction model assigned a masking risk above the cut-off point. Discovery rates are defined as the proportion of masked tumors in the “high-risk” group
  4. bProportion of “high risk” classified patients in the total study population, using boosting-based prediction model
  5. cLogistic regression model with the clinical predictors age, BMI, previous breast surgery, menopausal and HRT status, and imaging technique
  6. dLogistic regression model with the same clinical predictors and additionally PMD predicted by a boosting regression model beforehand
  7. eLogistic regression model with the clinical predictors and the observed PMD