All Camera-ready Abstracts

The accurate identification of vertebral fractures is important in
diagnosing the early stages of osteoporosis. Good mean segmentation
accuracy has been previously obtained for vertebrae on lateral dual
x-ray absorptiometry (DXA) scans, by modelling the spine as a sequence
of overlapping triplets of vertebrae.  However the Active Appearance
Models (AAM) used for each triplet sub-model were prone to
occasionally converge to incorrect local minima, mainly in the case of
severe (grade 3) fractures. The robustness of the AAM search sequence
has been improved by two means: 1) ensuring a better starting solution
by using information on the approximate centre of each vertebra; 2)
including an alternative starting solution at each vertebral level to
represent a possible severe fracture. The best of a set of normal and
fractured alternative sub-model solutions was selected at each
iteration. This provides a robust and accurate search method, even
with multiple severe fractures present. A mean segmentation accuracy
of 0.7mm was achieved on normal vertebrae, rising to 1.2mm on severely
fractured vertebrae.

In clinical studies of osteoarthritis using magnetic resonance
imaging, the placement of the test subject in the scanner tends to
vary and this can affect the outcome of automatic image analysis
methods for articular cartilage assessment, particularly in
multi-center studies. We have developed an automatic iterative
method that corrects for position variations by combining the two
steps: shifting the cartilage towards the expected position and
performing a voxel classification with the normalized position as a
feature. By applying this placement adjustment scheme to an
automatic knee cartilage segmentation method we show that the
inter-scan reproducibility is much improved and is now as good as
that of a highly trained radiologist.

In osteoarthritis (OA) the articular cartilage degenerates, thereby
losing its structure and integrity. Curvature analysis of the
cartilage surfaces has been suggested as a potential disease marker
for OA but until now there has been few results to support that
suggestion. We present two methods for surface curvature analysis,
one that estimates curvature on lower scales using mean curvature
flow, and one method based on normal directions of and distances
between surface points given by a cartilage shape model for high
scale curvature estimates. We show that both methods can distinguish
between healthy and osteoarthritic groups, and the shape model based
method can even distinguish healthy from mild OA with high
reproducibility, indicating the potential of surface curvature in
becoming powerful new disease marker for cartilage degeneration.

MRI cartilage measurement techniques need to be sufficiently
sensitive to detect small, focal changes if they are to be used as
biomarkers for OA drug development. Detailed cartilage thickness
maps were constructed from MRI's of 19 healthy female volunteers.
Anatomical correspondence between the volunteers was achieved by
constructing optimal statistical shape models of the bones.
Cartilage coverage across the cohort was used to define the region
of pre-morbid sub-chondral bone and trimming boundaries which
excluded the edges of the cartilage sheets. Functional sub-regions
of the joint were drawn on the mean bone shapes. The regions of
interest were propagated to all individuals, in an anatomically
consistent manner, using the model-based correspondences. Mean
cartilage thickness was measured within each region of interest.
Excluding the edges of the cartilage sheet from the mean thickness
measures increased the reproducibility, and hence sensitivity, of
the measures.

In this paper a method for automatic detection of erosive destructions
caused by rheumatoid arthritis is proposed. Based on hand radiographs
the algorithm detects erosions by means of an appearance model learned
from a training set of bones. The model is utilized to classify the
texture of the bone in the vicinity of the contour with respect to
erosive destructions. Quantitative results of the algorithm are
reported for a set of 17 radiographs of moderately and mildly diseased
hands.

A novel method for vertebral fracture quantification from X-ray images is
presented. Using
pairwise conditional shape models trained on a set of healthy spines, the most
likely normal
vertebra shapes are estimated conditional on all other vertebrae in the image.
The differences
between the true shape and the reconstructed normal shape is subsequently used
as a measure of
abnormality. In contrast with the current (semi-)quantitative grading
strategies this method takes
the full shape into account, it uses a patient-specific reference by combining
population-based
information on biological variation in vertebra shape and vertebra
interrelations, and it provides
a continuous measure of deformity.

The method is demonstrated on 212 lateral spine radiographs with in total 78
fractures. The
distance between prediction and true shape is 1.0 mm for unfractured vertebrae
and 3.7 mm for
fractures, which makes it possible to diagnose and assess the severity of a
fracture.

Cartilage thickness is a central indicator for monitoring osteoarthritis (OA)
progression. We present a novel, automatic method for quantification of
cartilage thickness from magnetic resonance imaging. First, an automatic voxel
classification produces a binary segmentation of the cartilage sheet. Second,
a statistical shape model is deformed into the binary segmentation. Finally,
the thickness map is extracted from the shape model.
  
We evaluate the cartilage thickness quantification on a collection of knee
scans with both healthy and OA subjects. The method shows high reproducibility
and proves to be able to separate healthy from OA subjects.

This paper considers the problem of automatic classification of textured
tissues in 3D MRI. More specifically, it aims at validating the use of
features extracted from the phase of the MR signal to improve texture
discrimination in bone segmentation. This extra information provides better
segmentation, compared to only using magnitude features. We also present a
novel multiscale scheme to improve the speed of pixelwise based classification
algorithm, such as support vector machines. This algorithm dramatically
increases the speed of the segmentation process by an order of magnitude
through a reduction of the number of pixels that needs to be classified in the
image.

Osteoarthritis (OA) is a degenerative joint disease that involves the wearing
down of the articular cartilage. A typical problem
has been quantifying progression and early detection of the disease. In this
study we develop a fully automatic method for investigating knee cartilage
homogeneity on 114 manually and automatically segmented T1 knee MR images from
subjects with no, mild or severe OA symptoms. To measure homogeneity we
characterize the tibial and femoral compart ments in each cartilage by several
statistical measures and then evaluate their ability to quantify OA
progression. The discriminatory power of each measure for separating the group
of healthy subjects from group of subjects having OA is tested statistically.
Our method outperforms a standard measure like volume in separating healthy
subjects from subjects having OA. We show that our method is reproducible
through a scan-rescan evaluation from additional 31 MR images.

Objective evaluation of structural changes in osteoarthritis is essential for
diagnosis and evaluation of disease progression. Since conventional
radiography is still the gold standard, in the present study a newly developed
digital method to analyze standard radiographs of knees was evaluated. Joint
space width (JSW), osteophyte area, subchondral sclerosis, deviation of the
angle of the joint, and eminentia height were measured using the interactive
application Knee Images Digital Analysis (KIDA) on a standard PC. Enlargements
on screen can be performed, when required. The application provides multiple
measures for all parameters as continue variables. Two observers evaluated the
radiographs on two different occasions with one-week interval. The observers
were blinded to the source of the radiographs and their previous measurements.
Intra- and interobserver variation was evaluated. 
The results demonstrate KIDA to be a reliable method to quantify and document
(for follow-up) the radiographic parameters of knee osteoarthritis. 

A fully automated method for analysis of Dynamic Contrast Enhanced MRI scans
of the metacarpophalangeal joints of the hand and assessing the extent and
magnitude of inflammatory activity in rheumatoid arthritis is put forward. The
method incorporates automated segmentation, spatial registration, and a
modelling of the underlying physical procedure. It affords robust and
consistent quantitation of the spatial and temporal properties of 4D datasets,
allows a more robust and consistent extraction of various parameters, and
provides information on procedure completion, hitherto unavailable, that is of
value in guiding the procedure and
informing the reliability of the parameter estimates.
The technique is demonstrated on 10 DEMRI studies and has potential for wider
application.

We introduce two new metrics for evaluation of segmentation
outputs obtained from Dynamic Contrast-Enhanced MRI data.
Considering a live application area, we demonstrate the shortcomings of
currently accepted algorithms.  We present a new supervised approach
as an enhanced derivation of a state-of-the-art method, and a novel
unsupervised approach, which enables automation of segmentation
output assessment. The proposed discrepancy measure considers
local blur, partial volume effects, intensity variations,
subtle contrast changes and the inconsistency of human experts.
We consider the approaches presented to be an
improvement on those prevailing, and worthy of wider experiment and
application.

The bicipital groove (BG) of the proximal humerus retains the tendon of the
long head of the biceps.  It is understood that the shape of the BG is related
to the probability of injury to the long biceps tendon (LBT).  Measurements
taken of the BG in previous studies from dry bones and radiographs (henceforth
classical measurements) are of single cross sections of the humerus, and may
therefore overlook important BG shape characteristics.  In this study, we test
the hypothesis that a novel, medial axis-based 3D shape descriptor captures
all relevant features measured in previous work, plus more.  To this end, we
review previous studies wherein classical measurements have been taken on
large numbers of BGs, yielding a distribution that reveals the nature of a
normal BG.  We develop an automated approach to replicating those measurements
on MRI to determine, for each of our data sets, the deviation from the mean of
all the classical measurements.  We train a classifier by pairing our 3D
representations with these deviations to evaluate the potential for computer
aided diagnosis of BG pathology based on our 3D shape descriptor.

Cartilage volume is the most obvious quantification of cartilage breakdown
when monitoring osteoarthritis progression. We present a novel algorithm for
speeding up voxel classification by an order of magnitude. This new
classification scheme is used for fully automatic segmentation of tibial and
femoral articular knee cartilage. 
  
We evaluate the method on a collection of 114+31+25 knee MR scans of both
healthy and OA subjects, and show that the segmentations are identical to the
segmentation from a straight-forward voxel classification method. Furthermore,
the method shows high reproducibility and proves to be able to separate
healthy from OA subjects.

Pathology of the supraspinatus muscle can involve tearing, which often leads
to atrophy and/or retraction of the muscle.  Retraction can be corrected
through a pull forward operation in surgery, whereas atrophy is generally not
correctable.  It is therefore important to distinguish between retraction and
atrophy.  However, since both of these conditions are characterized by a
reduction in size, we put forth a pilot study examining changes in 3D shape as
they relate to pathological conditions.  After segmenting the supraspinatus
muscle surface from MRIs representing 57 patients, we compute several
different 3D shape measures of the surfaces, and conclude that there are
statistically significant differences in shape and size between pathology
groups.

Abstract

 The objective of the present study is to create a dynamic 3D knee model which
represents tibio-femoral joint surfaces, bones and ligaments by consideration
of their geometric and material properties to simulate 0°-90° passive knee
flexion. Tibial plateaus of the tibia and condyles of femur are modeled as
ellipsoids as described in the literature. The contact forces between tibia
and femur are defined as frictionless mathematical model. Anterior, posterior
cruciate ligaments, medial, lateral collateral ligaments are represented as
non linear elastic springs. Knee flexion with and without internal-external
torque are simulated, and the results are compared with the literature for
slopped and flattened medial tibial plateau models. As a result, normal
internal rotation of tibia and adduction ranges are achieved for unloaded
condition in flattened model, but the knee flexion with forced
internal/external rotation are out of normal range for both models. 

Keywords: Knee, kinematics, anatomical dynamic modeling