Comparison of manual and automated lesion segmentation of digital breast tomosynthesis data - first experiments

In the past years, Digital Breast Tomosynthesis (DBT) has been established as a new and promising imaging method for a volumetric assessment of breast images with the advantage, that possible mass lesions are not obscured by overlay projections. In order to include DBT data in tomorrows Computer Assisted Diagnosis (CADx) Systems, these have to be extended accordingly. Within this contribution, we will shortly describe the necessary extensions for such CADx systems. Furthermore we will focus on building an annotated DBT image collection for research on CADx-systems, and we will provide initial segmentation results, based on two state-of-art methods, fuzzy C-means and region growing. These initial semiautomatic results as well as a combined ground truth obtained from a group of junior radiologists are compared to an expert s manual annotations.