Focal Analysis of Knee Articular Cartilage

Quantity and QualityDr. Tomos G. Williams

Imaging Science and Biomedical EngineeringUniversity of Manchester

Presentation for:

AstraZeneca Joint Imaging GroupAlderley Park

Thursday Sept. 16th 2004

Contents

Background Thickness mapping methodology

– Illustrated on CP77 Thickness mapping of CP78

– Preliminary Results for 9 patients Project Work Plan

– Disease Progression Hypotheses – Statistical Analyses

Deliverables

Image Acquisition and SegmentationCP77 CP78

CartilageT1 Fat-Suppressed

TOSCARegionGrowingat AZ

EndPointLiveWireat AZ

BoneT2

EndPointManualat ISBE

EndPointLiveWireat AZ(incomplete& poorquality)

50 60 70 80 90 100 110

70

80

90

100

110

120

130

140

150

X

AB121247_a2BFem

Y

Primary Analysis — Volume Measurement

No significant patterns during disease progression (OA16 Study)– Conflicting Literature on OA Cartilage

compartmental volume change Conclusions:

– Disease process more subtle• Involves swelling followed by

thinning– Need to detect focal changes

• Difficulties in defining cartilage edge in parallel planar segmentations

– Still need population trials• Changes too small to detect

in individuals Requirements:

– Aggregate, detailed thickness maps• Detect focal changes in a

population– Statistical analysis

Published Approaches Cohen, Z. A. et al.

– “Templates of the cartilage layers of the patellofemoral joint and their use in the assessment of osteoarthritic cartilage damage”. Osteo. and Cart., 11, 2003.

Kauffmann et al. – “Computer aided methods for

quantification of cartilage thickness and volume changes”. IEEE Trans. Biomed. Eng., 50(8), 2003.

Surface Alignment– Does not guarantee

anatomical equivalence– Problematic in

corresponding cartilages with lesions, especially on edges

Method Overview — CP77 as Illustration

Optimal Shape Optimal Shape ModelModel

CorrespondencCorrespondenceses

RegistratioRegistrationn

Cartilage Cartilage Thickness MapsThickness Maps

Aggregate Aggregate Thickness MapsThickness Maps

ImageImage

SegmentationsSegmentations

3D Surface3D Surface

BoneBoness

CartilageCartilagess

Identifying Correspondences

Use bone as a frame of reference– More stable than cartilage in longitudinal

studies– Consistent across population

Statistical Shape Model (SSM)– Set of corresponding points on each

example– A description of how these point vary

over the set of examples Optimal SSMs

– Correspondences that lie on anatomically equivalent points lead to simpler models

– Minimum Description Length approach to building optimal shape models

– Manipulate correspondences on each example to simplify the model

anatomically equivalent correspondences

Initial Model MDL: 4.297

Optimised MDL: 4.018

CP77 Bone Statistical Shape Models

N=19, One from each patient

Separate model for each bone compartment

Bone and Cartilage Surface Registration

Correct for:– Chemical shift artefact– Patient movement

• 10 minute scans Method:

– Register all 3 bone surfaces simultaneously

– Achieve consistent cortical bone thickness

SurfacesSurfacesCortical Bone Cortical Bone

ThicknessThickness

Unre

giste

red

Unre

giste

red

Registe

red

Registe

red

Measuring Cartilage Thickness

Define 3D normal to the bone surface

Individual cartilage thickness map

Inner Cartilage

Cartilage Thickness

Bone

Outer Cartilage

3D Normal

CP77 cartilage coverage

CP77 Cartilage ThicknessMean

Standard Deviation

80% Coverage Threshold

Thicker cartilage on load bearing regions

Consistent and Low Standard Deviation

CP77 Minimum/Maximum Thickness

Complete valid readings coverage

Statistical AnalysisLeft (reflected)

Right

Left - Right

P<0.002 that the difference observed was a chance effect

CP78 bone segmentation Lacking segmentations

for CP78 bones Augment CP77 Statistical

Shape Models with Image Intensity information – Active Appearance

Models Active Search of CP78

bone images– Automatic Segmentation

of bone structures– Automatic identification

of CP77 correspondence points

Fully implemented within EndPoint

CP78 Preliminary Results (n=9)

Mean Thickness MapsBaseline

6 months

Coverage

CP78 Mean Difference Map (n=9)

Disease Progression Hypotheses

1. Anatomical– Simultaneous analysis of thickness change at each

and every correspondence point.

2. Abnormality– Assess thickness change in regions where cartilage

is thin or thick at baseline (in comparison with CP77).

3. Lesion proximity– Assess thickness changes in regions surrounding a

lesion in the cartilage at baseline.

4. Opposition– Asses thickness changes at locations which

articulate with regions that are abnormal at baseline.

Work Plan – Statistical Analyses

Test What? Large ROIe.g. Femoral Cartilage

Small Anatomical ROI e.g. Central Medial Tibial Plateau

Statistical Approach

VolumeAnalysis 1

Analysis 2Univariate tests“Has the cartilage volume changed?”

Overall Thickness Map

Analysis 4 Analysis 3Multivariate Tests“Has the pattern of cartilage thickness changed?

Thickness at each Location

Analysis 5 Analysis 6Multiple Comparison Tests“Where has the thickness changed?”

Deliverables Results

– CP77 Normal Thickness Range

– CP78 Disease Progression Maps

– Statistical Analysis Publications

1. Bone corresponded cartilage thickness mapping methodology (CP77 for illustration)

2. Changes in Cartilage Morphology in Longitudinal Population Studies (CP78)

3. Normal Range Cartilage Thickness Maps (CP77 results)

Software Tools– Identification of

corresponding points through SSM optimisation

– Bone AAM for automatic segmentation

– Visualisation of thickness data on average bone surfaces

– Propagation of correspondence points to allow comparison of cartilage thickness with normal range.

– Implemented within EndPoint package