Non-fiducial, shape-based registration of biological tissue

Three-dimensional reconstruction of biological specimens often requires the registration of successive sections of tissue. However, the introduction of fiducial markers is exceedingly difficult in some imaging modalities. We describe a fully automated registration method for use with transmission electron microscopy or other modalities that does not require the existence of fiducial markers. This system employs an error minimization approach with progressive refinement that converges to the locally optimal registration across all considered objects. When used in conjunction with other methods for coarse refinement, a globally optimal solution is produced. The method can also be customized to the characteristics of the tissue involved. Among many of the possibilities, it can register based on contour location or, more interestingly, on contour shape, which effectively minimizes the surface tension of the resulting reconstruction. In addition, weighting factors are available to give increased consideration to larger objects when appropriate.