Automated volumetric breast density derived by statistical model approach

Abstract

Interest is growing in the developing automated breast density measures because of its strong association with breast cancer risk. Although a number of automated methods to quantify mammographic and volumetric density appeared, they still have issues with accuracy and reproducibility; there is demand for developing new accurate and automated breast density estimation techniques. The purpose of this paper is to design and to test a new approach for automatically quantifying true volumetric fibroglandular tissue volumes from clinical screening full-field digital mammograms. The approach consists in building a statistical model using a training set of digital mammograms with known measures of percent fibroglandular tissue volume, breast volume and fibroglandular tissue volume calculated by phantom based calibration method. To derive these measures, we follow the standard procedure in machine learning: feature generation, feature selection, regression classification of outputs, final model building and testing. The correlation of features to known volumetric breast volumes was analyzed. In addition, the performance of models created from different groups of features were studied. By building a statistical model with 28 degrees of freedom, we achieved an R2=0.83 between the predicted and measured volumetric breast densities for the testing set of 2000 mammograms which were independent of the training set of 2000 images. © 2014 Springer International Publishing.