A semiautomatic three-dimensional segmentation method for disarticulation of bone structures on spiral computed tomography images.

The use of binary thresholding for segmenting bone structures on spiral computed tomography images is negatively influenced by partial volume effects (PVEs) induced by the image acquisition. PVE leads to mixed voxels, making the binary decision "bone" or "nonbone" a difficult one to take. As a result, two distinct bone structures that are close to each other will often appear to be connected by this method. A typical example consists of "acetabulum/femural head" pairs in the pelvic region. To separate them, a clinical user must interactively draw a disarticulation line. This procedure is time consuming (often interaction in 50 slices is needed) and leads to unsmooth visualization of the disarticulated areas (by three-dimensional [3D] rendering techniques). We developed a semiautomatic cutting algorithm that leads to smooth disarticulated surfaces and considerably decreases the amount of user interaction. A sheet detection operator is applied to automatically separate bone structures. Detected sheets are used as disarticulation lines. Postprocessing ensures that sheets not relevant for the application do not influence the resulting image. Our approach is encapsulated in an interactive segmentation environment based on thresholding and 3D connected-component labeling. Results are shown for pelvic region, wrist, and foot bone disarticulations.