Validation of a 3D hidden-Markov model for breast tissue segmentation and density estimation from MR and tomosynthesis images

Breast cancer is the most common non-skin cancer and the second leading cause of cancer-related death in women here in the United States. Mammography is the current standard clinical imaging modality for breast cancer screening and diagnosis, and mammographic breast density (i.e. the percentage of the entire breast volume that is taken up by dense glandular tissue) has been shown to be a biomarker well correlated with cancer risk. However, a mammogram is limited by its projective nature, and its quantitative abilities would likely be surpassed by 3D imaging modalities. This study plans to extract quantitative 3D breast tissue density information using a fully automatic probabilistic model trained on pre-segmented MRIs. This model “ground truth” was obtained from MRIs for 293 breasts by iterative threshold-based voxel value classification. Before model training/testing, all images were processed to optimize the available range of values for this breast tissue segmentation task. After training a 3D hidden Markov model (HMM) on 10 segmented “ground truth” MR images, our model was validated by segmenting the remaining 283 breasts. Initial training/testing of the HMM on MRIs matched density to thresholding within 5% for 99/283 breasts and 10% for 171/213 breasts. After the same task-based value optimization method was applied to digital breast tomosynthesis (“tomo”) images, the same trained HMM was tested to segment tomo volumes of subjects with at least one MRI for validation. HMM segmentation was qualitatively superior at the most cranial/caudal slices in MRIs and quantitatively superior when tested across modalities. Its robustness and ease of modification give the HMM great promise and potential for expansion to more novel 3D breast imaging modalities like breast tomo.