Automatic segmentation and analysis of Magnetic Resonance images of the knee bones and cartilages

Mr Jurgen Fripp (2008). Automatic segmentation and analysis of Magnetic Resonance images of the knee bones and cartilages PhD Thesis, School of Information Technol and Elec Engineering, The University of Queensland.

Attached Files (Some files may be inaccessible until you login with your UQ eSpace credentials)
Name Description MIMEType Size Downloads
n3354094_PhD_abstract.pdf abstract application/pdf 39.01KB 20
n3354094_PhD_totalthesis.pdf totalthesis application/pdf 53.84MB 26
Author Mr Jurgen Fripp
Thesis Title Automatic segmentation and analysis of Magnetic Resonance images of the knee bones and cartilages
School, Centre or Institute School of Information Technol and Elec Engineering
Institution The University of Queensland
Publication date 2008-12
Thesis type PhD Thesis
Supervisor Stuart Crozier
Sebastien Ourselin
Total pages 357
Subjects 290000 Engineering and Technology
Formatted abstract The aim of this research was the development and evaluation of a non-invasive tool
to allow the automatic, accurate and reproducible quantitative measurement of subtle
changes in knee cartilage tissue over time and across a patient population. Non-invasive
information was obtained by acquiring, segmenting and analyzing high resolution Magnetic
Resonance (MR) images of the whole knee joint.
The interest in knee cartilage is from Osteoarthritis (OA), a disease characterized
by changes in structure and degeneration of cartilage tissue. The X-ray radiography is
the standard imaging modality for diagnosis. However, cartilage tissue is not directly
imaged so imprecise surrogate measures are used. It is now generally accepted that
radiographs do not provide the sensitivity required to perform short term studies or drug
trials into OA. Magnetic Resonance (MR) imaging allows the non-invasive acquisition
of high resolution images of the whole knee joint, including the cartilages. MR has the
potential to perform accurate and repeatable quantitative measurements of the cartilages
and allow the detection of small changes over time. This is important as it will aid
the development and refinement of cartilage-dedicated therapeutic strategies, surgical
treatments and identification of promising drug targets.
Quantitative measurements require the cartilage to be segmented, a task whose
accuracy can significantly influence the error and reproducibility of the analysis. This is
critical as only small volume losses (≃ 5% a year) and sub-millimetre thickness changes
are expected. Due to the structure and morphology of the cartilages as well as the nature
of MR acquisition, obtaining accurate segmentations can be problematic and are usually
performed manually by trained professionals. The hypothesis investigated was that
accurate, reproducible quantitative measurements of cartilage tissue can be obtained
automatically by developing and using advanced image processing techniques and
mathematical models, particularly the use of trained models of shape and appearance.
This was investigated using MR images from two sources 1) previous and current
studies that were kindly provided by fellow researchers and collaborators and 2) acquisitions
of volunteers. The MR images from different sources were acquired using different
MR sequences, parameters and field strengths. The images had a mixed demographic
of participants (male or female and young adult to elderly) who had healthy knee joints
or had mild to moderate OA. In some cases multiple scans were acquired from the same
patient, either on the same day or longitudinally to allow reproducibility and test-retest
experiments to be performed.A hybrid segmentation scheme was developed to process the MR images. This involved
the initial estimation of the bone location, accurate segmentation of the bones,
extraction of the bone-cartilage interface, estimation of cartilage properties and then
segmentation of the cartilages. The initial location of the bones was estimated using a
robust affine registration to an atlas, the bones were segmented using 3D active shape
models, while the bone-cartilage interface and cartilages were processed using priors,
statistical thickness models, image information and other constraints. Three other approaches
were also investigated, 1) an intensity non-rigid registration based on free form
deformation modelled using B-splines, 2) an improved watershed algorithm, and 3) a
tissue classifier that utilized bone presegmentations. These approaches were developed
and applied to the segmentation of individual magnitude MR images.
The acquisition and use of additional MR information can improve the accuracy of
segmentation algorithms. This was investigated using phase MR information, multiple
MR sequences and multiple echos. This additional information was utilized by training
classifiers to segment bone tissue using features extracted from the magnitude and/or
phase of the MR signal. This was extended to incorporate shape information to allow
more accurate and anatomically valid segmentations to be obtained.
Manual segmentations performed by experts were used as the ground truth, with
voxel based and surface based measures used to evaluate the segmentation quality of
the automatic segmentations. The Dice similarity coefficient (DSC) ranges from 0 to 1
and indicates the spatial overlap. The hybrid segmentation scheme obtained an median
DSC of (0.89, 0.96, 0.96) and (0.833, 0.826, 0.848) for the (patella, tibia, femur) bones
and cartilages. This was significantly better than the (0.810, 0.793, 0.849) and (0.732,
0.785, 0.758) obtained by the tissue classifier and non-rigid registration respectively. The
average DSC obtained for the whole cartilages by a semi-automatic watershed algorithm
(0.896) is slightly higher (0.891), however the hybrid approach is completely automatic
and obtains separate cartilage plates. Quantitative measures found a median volume
difference of (5.92, 4.65, 5.69)% and absolute thickness difference of (0.13, 0.24, 0.12) mm
for the (patellar, tibial, femoral) cartilages.
The use of additional information, in particular phase and magnitude information,
allowed more accurate bone classification results to be obtained with an average DSC
of 0.907 for the bone. The use of shape information further improved the robustness,
accuracy and anatomical validity of the segmentations, and obtained an overall DSC of
0.922. Further investigation is required to determine whether additional informationcan improve cartilage segmentation.
The innovative segmentation system developed allows the non-invasive automatic,
accurate and reproducible quantitative analysis of healthy and diseased cartilage tissue
from MR images. The presented scheme is competitive with semi-automatic methods,
with only a slight loss in accuracy (volume CV ≈ 5 − 6% compared to ≈ 2 − 3%), which
is more than made up by it being fully automatic, requiring no user interaction besides
a cursory visual validation of results.
Keyword Knee, Bones, Cartilages, Segmentation, Shape Modelling, Appearance Modelling, Magnetic Resonance Imaging, Osteoarthritis. xv

Citation counts: Google Scholar Search Google Scholar
Access Statistics: 299 Abstract Views, 46 File Downloads  -  Detailed Statistics
Created: Wed, 03 Dec 2008, 17:14:04 EST by Mr Jurgen Fripp on behalf of Library - Information Access Service