Abstract:
Creating a 3D reconstruction of an organ’s structure is a desirable resource for physiologists, researchers and students alike to better represent the structures of tissue as it would be in vivo. One of the most poorly understood organs in the human body is the uterus, especially during pregnancy. Major limitations to understanding the pregnant uterus are the rapid changes during development and the inaccessibility of direct measurements during pregnancy. However, a rare historical collection of gravid uterine specimens with the placenta in situ, the Boyd Collection, can provide us important insights that overcome some of these limitations. From the Boyd Collection, eight paraffin-embedded histological specimens from across gestation have been prepared as tissue slides and have only recently been digitised, however their historical preparation is not well suited for digital analysis. The primary aim of this thesis was to create 3D reconstructions from these 2D slides and perform segmentation of important tissue types.
Existing digital reconstruction tools proved inadequate to achieve the aim, so a custom software tool was developed. This tool performs significant pre-processing to isolate individual sections from their original tissue slides. Rigid and non-rigid registration methods were developed to create continuity of tissue structures. Interpolation between missing sections is performed to recreate an anatomically correct organ volume. A machine learning workflow was developed to take the outputs of the registration method and segment the volume of specimens into its important tissue types.
This thesis has produced 3D reconstruction of the eight specimens and produced an initial segmentation of one specimen. These virtual reconstructions allow for the 3D visualisation of the tissue, not previously possible in the 2D slides. The methods developed for the segmentation have provided a proof-of-concept on how to create 3D segmentations of the uterine tissue types for all specimens.