3D automatic breast boundary detection
Breast screening with an ultrasound tomography scanner  involves immersion of the breast in the water. The breast is scanned along the coronal axis from the chest wall to the nipple region. To improve the visualization of the generated 3D image, it is desirable to remove the water background. To this end, the 3D boundary of the breast must be accurately estimated. We developed an algorithm based on active contours that automatically detects the boundary of a breast using a 3D coronal stack of ultrasound attenuation images obtained from an ultrasound tomography scanner. Our approach builds upon a method described in  which uses parametric active contours that are represented with exponential B-splines. The authors in  demonstrate that their active contour approach can approximate any blob-like objects with good accuracy, and can perfectly reproduce spheres and ellipsoids. The optimization process is remarkably fast and good approximation can be obtained using only a few control points. The technique is appealing in that the smoothness of the object boundary is implicitly taken into account. We made a number of modification to the original technique such as: 1) mirroring of the input image stack around a chest wall; 2) adapting the region of interest; 3) making an iterative optimization algorithm that successively optimizes the boundary etc. As a result our algorithm is fully automated, fast, and robust to artifacts that may arise from the ultrasound tomography reconstruction process.
We demonstrate the effectiveness of the proposed technique using clinical data sets obtained from SoftVue scanner developed by Delphinus Medical Technologies.