The automatic segmentation of three-dimensional medical images into anatomically relevant structures is a fundamental bottleneck in timely presentat ion of three-dimensional therapeutic data sets. We present a novel technique for the automatic volumetric segmentation of medical images that relies on a ``shock-based'' representation of shape. Informally, ``bubbles'', or small spherical deformable structures, are randomly initialized as fourth-order shocks in the homogeneous areas of the three-dimensional image. These bubbles then grow, merge, split, shrink and in general deform under physically-motivated ``forces'', but slow down and come to a halt near differential structures, \EG, the surface of the ventricle cavity. The final outcome is a set of closed surfaces representing visible structures in three-dimensional initial data. The differential structure is represented by edge intensity gradient, and can also involve a variety of local measurements,such as tissue density and texture. Since like their two dimensional counterparts, three-dimensional bubbles allow for integration of a variety of local measurements, this technique allows for the registration and fusion of data from diverse image modalities. We have applied this process to a variety of these modalities including MR, CT, and US. Here we report our results for MRI and MRA images of brain, vascular structures are successfully segmented with substantial detail and clarity.