Automatic reconstruction from serial sections

In many experiments in biological and medical research, serial sectioning of biological material is the only way to reveal the three dimensional (3D) structure and function. For a number of reasons other 3D imaging techniques, such as CT, MRI, and confocal microscopy, are not always adequate because they cannot provide the necessary resolution or contrast, or because the specimen is too large, or because the staining techniques require sectioning. Therefore for the foreseeable future reconstruction from serial sections will remain the only method for 3D investigations in many biomedical fields. Reconstruction is a difficult problem due to the loss of 3D alignment as the sections are cut and, more seriously, the systematic and random distortion caused by the sectioning and preparation processes.Many authors have reported how serial sections can be registered by means of fiducial markers or otherwise, but there have been only a few studies of automated correction of the sectioning distortions. In this thesis solutions to the registration problem are reviewed and discussed, and a solution to the warping problem, based on image pro¬ cessing techniques and the finite element method (FEM), is presented. The aim of this project was to develop a fully automatic method of reconstruction in order to provide a 3D atlas of mouse development as part of a gene expression database. For this purpose it is not necessary to warp the object so that it is identical to the original object, but to correct local distortions in the sections in order to produce a smooth representative mouse embryo. Furthermore the use of fiducial markers was not possible because the reconstructions were from already sectioned material.In this thesis we demonstrate a new method for warping serial sections. The sections are warped by applying forces to each section, where each section is modelled as a thin elastic plate. The deformation forces are determined from correspondences between sections which are calculated by combining match strengths and positional information. The equilibrium state which represents the reconstructed 3D image is calculated using the finite element method. Results of the application of these methods to paraffin wax and resin embedded sections of the mouse embryo are presented