Autonomous Direct 3D Segmentation of Articular Knee Cartilage
Download
Report
Transcript Autonomous Direct 3D Segmentation of Articular Knee Cartilage
Autonomous Direct
3D Segmentation of
Articular Knee Cartilage
Author :Enrico Hinrichs, Brian C. Lovell,
Ben Appleton, Graham John Galloway
Source :Australian and New Zealand Intelligent
Information Systems, 10-12 December 1(1),
pages 417-420, Sydney
Speaker : Ren-LI Shen
Advisor : Ku-Yaw Chang
1
Outline
Introduction
Segmentation
Discussion and results
2
Introduction
Osteoarthritis (OA) occurs
◦ 30 to 70 years
◦ Years<30:High-impact sports player
Using MRI
◦ High-contrast cartilage images
Focus on
◦ Automation segmentation
◦ Improvement accuracy of cartilage
measurements
3
Introduction
Expected outcomes
◦ Autonomous segmentation method
◦ Early detection of Pathology-Associated
Changes
◦ Detection of early onset OA
Problem
◦ Can’t use only grey level features
Similar cartilage contact zones
Between the femoral and tibia cartilage
4
Introduction
5
Introduction
Solution of these drawbacks is the main
objective of this work
◦ Develop a fully automated 3D segmentation
◦ Non-linear diffusion(NLD)
Cartilage lesion classification system by
Outerbridge
6
Introduction
7
Outline
Introduction
Segmentation
Discussion and results
8
Segmentation
Previous work: B-Spline snakes
Develop a fully automated segmentation
method
◦ Using NLD and level sets
Articular cartilage is difficult to segment
◦ It is a thin structure (1-2mm)
Another difficulty
◦ Cannot be used to reliably cartilage degeneration
Multispectral Segmentation
Manual Segmentation
9
Segmentation
Non-Linear Diffusion
◦ Overcome meaningful details are removed as
less important details
◦ Enables image simplification
◦ Preserves large intensity discontinuities and
sharpens the edges of objects
10
Segmentation
Non-Linear Diffusion
I is the image at time t and c is the
diffusivity function
11
Segmentation
Algorithm Development Using 3D Level
Sets
◦ Cartilage surface S is represented in space R
3
◦ Three dimensional level set function φ maps
to one dimension R
12
Segmentation
Match the cartilage contour as a partial
differential level set equation
|∇φ | describes the normal velocity of the
surface
F defined range of surface deformations
◦ Match the cartilage contour
13
Outline
Introduction
Segmentation
Discussion and results
14
Discussion and results
Automatic segmentation
◦ Speed up drug development
◦ Improve OA medication
The algorithm development is currently in
the initial phase and more results will be
provided soon
15