Accurate Measurement of Cartilage Morphology Using a 3D Laser Scanner

1
Accurate Measurement of Cartilage Morphology
Using a 3D Laser Scanner

• Nhon H. Trinh1, Jonathan Lester2, Braden C.
  Fleming1,2,3, Glenn Tung2,3, and Benjamin Kimia1
• 1Division of Engineering, Brown University,
  Providence, RI 02912
• 2Department of Orthopedics, Brown Medical School,
  Providence, RI 02903
• 3Department of Diagnostic Imaging, Brown Medical
  School, Providence, RI 02903
• 4Rhode Island Hospital, Providence, RI 02903

2nd International Workshop onComputer Vision
Approaches to Medical Image Analysis (CVAMIA
2006) May 12, 2006
2
Why measure cartilage morphology?

• Osteoarthritis (OA)
• common and a leading cause of disability in
  elderly people.
• associated with degeneration of cartilage in
  articulating joints.
• To monitor OA use morphology measurements of
  cartilage, e.g. overall volume and thickness.

3
The challenges

• MRI is imaging modality of choice for knee
  cartilage
• Non-invasive
• Able to differentiate soft tissues
• Quantification challenges
• Typical voxel size 0.31.0 mm
• Average knee cartilage thickness 1.3-2.5 mm
• Change in thickness due to OA can be overcome by
  one pixel error ( 25).
• It is imperative to know the accuracy of
  morphology measurements.
• The need for ground truth data.

4
Obtaining ground truth of knee cartilage is
difficult

• Not possible to obtain the cartilage in one piece
• Thin curved structure, thickness typically less
  than 6mm. Thinner for OA patient.
• Strong bond with knee bones.
• The cartilage cannot be left outside for a long
  time
• its content is mostly water

http//www.uchospitals.edu
5
Existing methods

• 1. Water displacement of surgically removed
  cartilage tissue
• Only measure volume
• Prone to error and requires a highly skilled
  technician
• 2. High resolution scans of anatomical sections
  with high precision saws.
• Only measure thickness in the sectioning
  direction

6
Existing methods

• 3. Computed tomography (CT) arthrography
• Same resolution problem as MR images
• 4. Stereophotogrammetry
• Extensive work to calibrate cameras
• Specimen is attached to a calibration frame, thus
  limiting number of views
• 5. Laser scanner
• Major differences with the proposed method

7
3D Laser scanner

• Create a 3D point cloud sampling of the
  specimen's surface
• Highly accurate
• Established technology commercial products and
  technical support available
• Wide range of algorithms available

ShapeGrabber PLM 300 with scan head SG-1000
(depth resolution 5.0µm)
8
The approach

• The cartilage volume is the difference between
  two bone surfaces one with cartilage and one
  without cartilage.
• To find the cartilage volume, we find those two
  surfaces and subtract them.

9
Equipment

• Laser scanner
• Shape Grabber PLM300 with scan head SG-1000
  (Vitana Corporation, Ottawa, Ontario, Canada)
• Depth range 250-900 mm, depth accuracy 5.0 µm
• PolyWorks IMAlign and IMMerge
• software packages to process 3D point clouds

10
Specimen

• Cadavers
• 5 intact fresh frozen human cadavers from the
  right limb (age 51-59, 3 males/2 females)
• MR-scanned before dissected

Femur
Tibia
11
Specimen (contd)

• 2. Synthetic knee bone and cartilage model
• Synthetic knee cartilage constructed in our lab,
  firm and water-proof.
• Used to validate the proposed
• 16 fiducial points are marked on the cartilage
  surface for thickness validation.

Tibia
Femur
12
The plan

• Method
• Validation experiments with synthetic models
• Obtain ground truth of the cadavers cartilage

13
Method
Scan the bone surface with cartilage intact
Dissolve cartilage off the bone
Scan the bone surface without cartilage
Reconstruct two bone surfaces
Reconstruct cartilage surface mesh
Compute cartilage morphology from the mesh
14
Method
Scan the bone surface with cartilage intact
Dissolve cartilage off the bone
Scan the bone surface without cartilage
Reconstruct two bone surfaces
Reconstruct cartilage surface mesh
Compute cartilage morphology from the mesh
15
Scanning the bone surface

• Scanning procedure
• 20 scans for each bone.
• Time 1 hour/bone.
• Properties
• Easy to set up
• At least 30 overlap among adjacent scans.
• Redundant coverage of the cartilage.

16
Example
17
Method
Scan the bone surface with cartilage intact
Dissolve cartilage off the bone
Scan the bone surface without cartilage
Reconstruct two bone surfaces
Reconstruct cartilage surface mesh
Compute cartilage morphology from the mesh
18
Dissolving cartilage off the bone

• Immerse bones in Clorox bleach 5.25 sodium
  hypochlorite.
• Tibia 4-5 hours
• Femur
• Regularly rotated to prevent the bleach from
  dissolving the bone and soft tissue
• Time 8-9 hours

19
Method
Scan the bone surface with cartilage intact
Dissolve cartilage off the bone
Scan the bone surface without cartilage
Reconstruct two bone surfaces
Reconstruct cartilage surface mesh
Compute cartilage morphology from the mesh
20
Reconstruct bone surfaces
Align the range images
Merge the aligned images
Smooth and fix topology problems
PolyMender (Ju SIGGRAPH 2004)

• PolyWorks IMAlign
• manual aligment interface
• ICP

• PolyWorks IMMerge
• handle outliers

21
Surface reconstruction example
22
Method
Scan the bone surface with cartilage intact
Dissolve cartilage off the bone
Scan the bone surface without cartilage
Reconstruct two bone surfaces
Reconstruct cartilage surface mesh
Compute cartilage morphology from the mesh
23
Reconstruct the cartilage volume

• Goal a triangular mesh of the knee cartilage
  from the two bone surfaces.
• Procedure
• Align the two surfaces and construct an error map

The two reconstructed bone surfaces do not
overlap completely on the bone body.
24
Reconstruct the cartilage volume

• Procedure (contd)
• Manually outlines the cartilage area.
• Project the outline orthogonally to both bone
  surfaces to segment the cartilage regions.

25
Reconstruct the cartilage volume

• Procedure (contd)
• Connect the two segmented surface patches with a
  band-like triangular mesh

26
Cartilage reconstruction examples
27
Method
Scan the bone surface with cartilage intact
Dissolve cartilage off the bone
Scan the bone surface without cartilage
Reconstruct two bone surfaces
Reconstruct cartilage surface mesh
Compute cartilage morphology from the mesh
28
Quantify knee cartilage morphology

• Volume

Vertices
Faces
29
Quantify knee cartilage morphology

• Thickness
• Use outer surface as reference surface and
  compute closest point on the inner surface.
• Algorithm spatial partitioning algorithm by
  Aspert et al. (ICME 2002)

30
Validation experiments

• Synthetic cartilage models
• Scanned as if they were from real cadavers.
• Reconstruct a triangular mesh of the cartilage
• Compute volume and thickness at fiducial points.
• Compare results with ground truth obtained from
  current standard methods
• volume using water displacement
• thickness using a caliper

31
Volume measurements

• Synthetic volume measurements within accuracy
  range
• Average discrepancy 5

32
Thickness measurements

• Laser-scanner thickness measurements are within
  error range of calipers.
• Discrepancy less than 5 error on average

33
Work in progress

• Overall goal Validate segmentation algorithms
  on MR images of the cadavers
• Volume estimate from manual segmentation

• Average discrepancy 14.7
• Using measurements from laser-scanning method as
  ground truth is appropriate.

34
Related work Koo et al. Osteoarthritis and
Cartilage 2005

• Use a 3D laser scanner to scan femurs of porcine
  knees (with and without cartilage)
• Align using the attached frame
• Construct thickness map
• Advantages of our method
• Construct a 3D mesh instead of thickness map
• Specimen not attached to a frame
• Provide validation

35
Summary

• A method using a 3D laser scanner to reconstruct
  the triangular mesh of the knee cartilage.
• Able to measure multiple morphological properties
• Validation using synthetic knee bone and
  cartilage models.
• Can be used to validate MR measurements.

Future directions

• Automate the cartilage boundary outlining
  process.
• Validate for accuracy using a more reliable
  methods, e.g. coordinate measuring machine.
• Validate the method for reproducibility

36
Acknowledgements

• NSF grant IIS-0413215
• NIH grant AR047910S1
• Professor Richard Fishman, Visual Arts
  Department, Brown University

37
Thank you