Towards automated three-dimensional tracking of nephrons through stacked histological image sets
Files
Date
2016-03-17
Authors
Bhikha, Charita
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The three-dimensional microarchitecture of the mammalian kidney is of keen
interest in the fields of cell biology and biomedical engineering as it plays a
crucial role in renal function. This study presents a novel approach to the
automatic tracking of individual nephrons through three-dimensional histological
image sets of mouse and rat kidneys. The image database forms part of a previous
study carried out at the University of Aarhus, Denmark. The previous study
involved manually tracking a few hundred nephrons through the image sets in
order to explore the renal microarchitecture, the results of which forms the gold
standard for this study. The purpose of the current research is to develop methods
which contribute towards creating an automated, intelligent system as a standard
tool for such image sets. This would reduce the excessive time and human effort
previously required for the tracking task, enabling a larger sample of nephrons to
be tracked. It would also be desirable, in future, to explore the renal
microstructure of various species and diseased specimens.
The developed algorithm is robust, able to isolate closely packed nephrons
and track their convoluted paths despite a number of non-ideal conditions such
as local image distortions, artefacts and connective tissue interference. The
system consists of initial image pre-processing steps such as background removal,
adaptive histogram equalisation and image segmentation. A feature extraction
stage achieves data abstraction and information concentration by extracting shape
iii
descriptors, radial shape profiles and key coordinates for each nephron crosssection.
A custom graph-based tracking algorithm is implemented to track the
nephrons using the extracted coordinates. A rule-base and machine learning
algorithms including an Artificial Neural Network and Support Vector Machine
are used to evaluate the shape features and other information to validate the
algorithm’s results through each of its iterations.
The validation steps prove to be highly effective in rejecting incorrect tracking
moves, with the rule-base having greater than 90% accuracy and the Artificial
Neural Network and Support Vector Machine both producing 93% classification
accuracies. Comparison of a selection of automatically and manually tracked
nephrons yielded results of 95% accuracy and 98% tracking extent for the
proximal convoluted tubule, proximal straight tubule and ascending thick limb of
the loop of Henle. The ascending and descending thin limbs of the loop of Henle
pose a challenge, having low accuracy and low tracking extent due to the low
resolution, narrow diameter and high density of cross-sections in the inner
medulla. Limited manual intervention is proposed as a solution to these
limitations, enabling full nephron paths to be obtained with an average of 17
manual corrections per mouse nephron and 58 manual corrections per rat nephron.
The developed semi-automatic system saves a considerable amount of time and
effort in comparison with the manual task. Furthermore, the developed
methodology forms a foundation for future development towards a fully
automated tracking system for nephrons.
Description
A dissertation submitted to the Faculty of Engineering and the Built Environment,
University of Witwatersrand for the degree of Master of Science in Engineering.
August, 2015