Use this URL to cite or link to this record in EThOS:
Title: Novel image processing methods for cellular micrographs
Author: Nam , David Clarence
ISNI:       0000 0004 5920 6759
Awarding Body: University of Bristol
Current Institution: University of Bristol
Date of Award: 2014
Availability of Full Text:
Access from EThOS:
Interdisciplinary research plays an ever-increasing role in advancing the state-of-the art in bioimage analysis. Advances in biomedical imaging will give biologists the tools to analyse large numbers of electron and light microscopy images, quickly and accurately. Currently, many biologists analyse images manually. This is not reproducible and susceptible to subjective bias. Image segmentation and registration are two fundamental techniques, which are frequently applied to microscopy images. Cellular images contain many complex structures, and to be able to get accurate information from these images, standard segmentation and registration techniques cannot be applied. This thesis presents new. approaches to automatically analyse light and electron microscopy images. We first introduce a method to segment and measure insulin granule cores and membranes, from transmission electron microscopy images of beta cells of rat islets of Langerhans. The algorithm proceeds through two critical steps, firstly core segmentation and then membrane segmentation. For core segmentation we do an initial segmentation using a novel level-set active contour. Post-processing is done to remove over segmentation, from other organelles and particles within the cytoplasm. A final refining step is done using a novel dual level-set active contour. Membrane segmentation uses the initial segmentation, from core segmentation, as one of its inputs. Membranes are sampled and scaled, then gaps in the membrane are filled. It is possible that the granule membrane is at the core, so we introduce a novel convergence filter to verify our segmentation. We validate our method by doing comparisons against manual segmentation as well against the state-of-the-art. Our results also compare favourably to previously published data. We then propose a new algorithm for granule segmentation in 3-D transmission electron microscopy. Due to differences in sample preparation, the 3-D images have poor contrast. This presents a challenge for membrane segmentation. We propose a novel region-based active surface in a Bayesian framework with a core boundary prior, to overcome this. In some cases the granule membrane is not visible and is only characterized by a halo around the core. Our novel active surface shows promising results on our data set. Towards the end of the thesis we focus on a feature-based registration algorithm for largely misaligned bright-field light microscopy images and transmission electron microscopy images. We first detect cell centroids, using a gradient-based single-pass voting algorithm. Images are then aligned by finding the flip, translation and rotation parameters, which maximizes the overlap between pseudo-cell-centres. We demonstrate the effectiveness of our method, by comparing it to manually aligned images. Combining registered light and electron microscopy images together can reveal details about cellular structure with spatial and high-resolution information.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available