[PDF] - Shape Modelling Aquisition Reconstruction Processing 17-06-2009 PDF Document

SLIDE 1

1

Shape data modelling and analysis as a support in the diagnosis of vascular diseases

M. Attene, M. Mortara, G. Patanè

CNR-IMATI Ge Speaker: Michela Mortara michela@ge.imati.cnr.it

17-06-2009 Workshop on Anatomical Models, INRIA 2

Shape Modelling

Aquisition
Reconstruction
Processing
…

SLIDE 2

2

17-06-2009 Workshop on Anatomical Models, INRIA 3

Shape Analysis

Characterization
Segmentation
Structuring
…

17-06-2009 Workshop on Anatomical Models, INRIA 4

In the following…

Tailor
Plumber
Convex Approximation
ShapeAnnotator

with possible applications in the medical field

SLIDE 3

3

17-06-2009 Workshop on Anatomical Models, INRIA 5

Tailor

Use a curvature analysis to compute the

main features of a given shape and organize them into an abstract representation

Automate processes like

matching, retrieval, comparison

f shapes.

M.Mortara, G.Patané, M.Spagnuolo, B.Falcidieno, J.Rossignac "Blowing Bubbles for the multiscale decomposition of triangle meshes" Algorithmica - special issue on shape algorithms. Vol 38, (1), pp. 227-248, Springer, 2003.

17-06-2009 Workshop on Anatomical Models, INRIA 6

Tailor

Label of v at scale i ←

configuration of the intersection between the mesh and a sphere having radius Ri and centred in v.

Curvature Topology Geometric attributes

SLIDE 4

4

17-06-2009 Workshop on Anatomical Models, INRIA 7

Classification criteria

Number of connected components:
1,2, 3 or more

17-06-2009 Workshop on Anatomical Models, INRIA 8

Classification criteria

One intersection curve
Curvature characterization:
Sharp
Rounded
Blend

SLIDE 5

5

17-06-2009 Workshop on Anatomical Models, INRIA 9

GC(v)=L/R ∈[0, +∞)
On a plane: GC(v)=2π
On a spike: GC(v)< 2π →0
On a blend: GC(v)>2π → +∞
No distinction between

convex and concave

R R α l P l : α = 2πR : 2π α = l / R Σ αi = L / R

Multi-scale curvature on meshes

17-06-2009 Workshop on Anatomical Models, INRIA 10

Classification criteria

Two intersection curves
Relative length characterization:
Cylindrical Lmax≤ 2 Lmin
Conic otherwise

SLIDE 6

6

17-06-2009 Workshop on Anatomical Models, INRIA 11

Classification criteria

Status:
One int.
Convex
Concave
More int.
Full
Empty

17-06-2009 Workshop on Anatomical Models, INRIA 12

Final classification

HOLLOW SPLIT WELL FUNNEL JOINT LIMB BLEND DIP PIT MOUNT TIP

Tip Tip Tip Tip Cylindrical Cylindrical Saddle Saddle Conic Conic Smooth Smooth Cylindrical Cylindrical Concave Concave Branching Branching

SLIDE 7

7

17-06-2009 Workshop on Anatomical Models, INRIA 13

Query Language

Each vertex is

described by a vector of labels

The ith label

corresponds to the vertex characterization at scale Ri

17-06-2009 Workshop on Anatomical Models, INRIA 15

Multi-scale Morphological analysis of the cortex

Tailor results

SLIDE 8

8

17-06-2009 Workshop on Anatomical Models, INRIA 16

Curvature-based Skeleton

17-06-2009 Workshop on Anatomical Models, INRIA 17

ProTailor

Parallel implementation on a Linux Beowulf Cluster of

16 PCs

2.66 GHz Pentium IV processor
1 GB of Ram
2 EIDE 80 GB disks interfaced in RAID 0
by Antonella Galizia,IMATI
The scalability of the

algorithm is almost linear (molecule model, 35MB, 31 minutes -> 2 minutes)

M.Mortara, A.Galizia “ProTailor: a paralle operator for extremely fast shape analysis in Bioinformatics applications” in Proceedings of PDP2009, special session on Bioinformatics applications.

SLIDE 9

9

17-06-2009 Workshop on Anatomical Models, INRIA 18

Tailor on volumes

on tetrahedral meshes, computes the sphere

surface inside the volume

17-06-2009 Workshop on Anatomical Models, INRIA 19

Plumber

Segmentation into tubular features and

“bodies”

Is based on the Tailor characterization
Works in a multi-scale fashion wrt tube

section size

Computes axis and sections of each

tubular feature

SLIDE 10

10

17-06-2009 Workshop on Anatomical Models, INRIA 20

Plumber

The algorithm works in the following steps:

Selection of the scale R
Classification of vertices and

identification of seed limb region

Tubular feature extraction
Increase R and repeat
M. Mortara, G. Patané, M. Spagnuolo, B. Falcidieno, and J. Rossignac. Plumber: A Multi-

scale Decomposition of 3D Shapes into Tubular Primitives and Bodies, Proc. of Solid Modeling and Applications, 2004

17-06-2009 Workshop on Anatomical Models, INRIA 21

Plumber

A sphere is centred

in the barycentre of the medial loop.

The sphere rolls in the two possible

directions until a halting condition occurs.

SLIDE 11

11

17-06-2009 Workshop on Anatomical Models, INRIA 22

Plumber

At each step, the sphere generates a new

cross section and a new axis point.

Stop when:
The tube ends
Curve length over threshold
Bifurcation

17-06-2009 Workshop on Anatomical Models, INRIA 23

Results

SLIDE 12

12

17-06-2009 Workshop on Anatomical Models, INRIA 24

Results

17-06-2009 Workshop on Anatomical Models, INRIA 25

Plumber on Point Clouds

SLIDE 13

13

17-06-2009 Workshop on Anatomical Models, INRIA 26

Identification of human body parts

17-06-2009 Workshop on Anatomical Models, INRIA 27

Generalization of the hierarchical face clustering

(Garland et al. 01)

Automatic generation of a binary tree of clusters,

each fitted to one of the available primitives: planes, spheres and cylinders

At the first step, each triangle is

considered to be a cluster

Then, adjacent triangles are

clustered according to cluster-to-primitive fitting

M. Attene, B. Falcidieno, and M. Spagnuolo. Hierarchical Mesh Segmentation based on

Fitting Primitives. The Visual Computer, 22, 2006

Fitting Primitives

SLIDE 14

14

17-06-2009 Workshop on Anatomical Models, INRIA 28

Segments a shape into a hierarchy of nearly

convex parts

Works on tetrahedra
Applications:
fast region selection

from complex models

shape segmentation
shape approximation
deformation, editing

Hierarchical Convex Approximation

Marco Attene, Michela Mortara, Michela Spagnuolo and Bianca Falcidieno “Hierarchical Convex Approximation

f 3D Shapes for Fast Region Selection” Computer Graphics Forum, Vol. 27, No. 5 (SGP'08 Procs.), pp. 1323-

1333, 2008

17-06-2009 Workshop on Anatomical Models, INRIA 30

Region selection

The tree of clusters can be traversed

SLIDE 15

15

17-06-2009 Workshop on Anatomical Models, INRIA 31

Complex Selections

Tracking cutting

lines or surface strokes can be complicated

Difficult topology
Occlusions

Here the heart is connected with multiple vessels and tissues, and is tightly occluded by the chest, thus selecting it through cutting lines would be rather unpractical. Through our mechanism the selection required just a mouse click and a wheel rotation.

17-06-2009 Workshop on Anatomical Models, INRIA 35

Annotation

Psychological surveys show that humans “understand”

shapes by recognizing interesting sub-parts and their structure (Marr 1982, Biederman 1987).

Shape Segmentation
For specific contexts, it is possible to describe what these

“features” are and how they are structured.

Geometric Description of the features
Structural Description of the shape
Semantic Annotation

cylinder plane Cylinder IN plane Through Hole

SLIDE 16

16

17-06-2009 Workshop on Anatomical Models, INRIA 36

ShapeAnnotator

Shape

A surface mesh

Domain

An OWL ontology

CG Tools

Segmentation Plug-ins

Shape Annotator Expert Expert

Knowledge Base

Abstracted Shape

A segmented mesh

Instance Instance

User knowledge → usable explicit content

17-06-2009 Workshop on Anatomical Models, INRIA 37

ShapeAnnotator

The ShapeAnnotator is an open-source software

project hosted by sourceforge.net

http://shapeannotator.sourceforge.net

SLIDE 17

17

17-06-2009 Workshop on Anatomical Models, INRIA 38

Scenario

Aquisition/reconstruction Mesh simplification Part selection Annotation Analysis Search/retrieval Comparison Sharing … … … … … …

17-06-2009 Workshop on Anatomical Models, INRIA 39

Monitoring shape changes over time
Comparison of the shape of anatomical regions

among patients

Semantic rendering of anatomical regions and

surgery planning

Analysis, automatic extraction of metadata,

automatic or supported annotation of anatomical regions for future retrieval, comparison and analysis

f collected data.