Three-dimensional reconstruction: methods of improving image registration and interpretation.

Abstract

(i) Image registration. The use of serial images for computerised three-dimensional reconstruction necessitates the inclusion of three separate sources of information at the stage of data input. These are (i) artificial registration points or fiducials, (ii) a calibration scale and (iii) an outline of each slab of the section to be included in the reconstruction. Most traditional methods rely on the production of drawings of the contours of the structure under investigation which also include both registration points and a calibration scale. We report on a method which considerably reduces the time involved at this labour intensive stage of reconstruction and in addition allows subsequent reconstructions of different structures to be performed without new drawings. Use is made of computerised alignment of tissue sections and the production of composite photomicrographs of both the tissue under investigation and an accurately registered stage micrometer scale. (ii) Improving image interpretation. Images derived from computerised three-dimensional reconstruction can be affected by the number of coordinates used to form the contour of each slice of a structure and by the number of slices that are used to construct the final model. Too little or too much data can considerably reduce the ability of the observer to interpret accurately the image generated by the computer. We report on a feature-based method which enables the experimenter to assess objectively the amount of data required in the two-dimensional plane, i.e. the number of data points per slice, and the three-dimensional plane, i.e. number of slices per structure, so that optimal reconstructions are generated.