Segmentation of breast region from MR images using multi-state cellular neural networks

Computer assisted evaluation of magnetic resonance (MR) images for breast density assessment or lesion localization requires accurate separation of breast tissues from other tissues and regions of the body, such as the chest muscle, lungs, heart and ribs. In this study, we introduce a semi-automated algorithm that segments breast region from non fat-suppressed T2-weighted axial breast MR images. The algorithm employs three specially designed multi-state cellular neural networks (CNNs) connected in cascade. Analysis of 106 high-resolution images from 23 women acquired using a 3T MR scanner shows that the algorithm is exceptionally effective with high precision, high true-positive volume fraction, and low false-positive volume fraction with an overall performance of 99.1±2.0%, 99.4±1.4%, and 0.1±0.2%, respectively. The use of multi-state CNN reduces the false segmentations on the images due to noise, intensity inhomogeneity and partial volume artifacts.