Segmentation of the left atrial appendage from 3D images

41
Segmentation of the left atrial appendage from 3D images Pol Grasland-Mongrain 20/04/2009 – 28/08/2009

description

Segmentation of the left atrial appendage from 3D images. Pol Grasland-Mongrain 20/04/2009 – 28/08/2009. Views of the Left Atrial Appendage. Views of the Left Atrial Appendage. Variable shapes 1 to 19 cm 3 Function ? Has to be ablated sometimes. Motivation. Current implementation: - PowerPoint PPT Presentation

Transcript of Segmentation of the left atrial appendage from 3D images

Page 1: Segmentation of the left  atrial  appendage from 3D images

Segmentation of the left atrial appendage from 3D images

Pol Grasland-Mongrain20/04/2009 – 28/08/2009

Page 2: Segmentation of the left  atrial  appendage from 3D images

Views of the Left Atrial Appendage

Page 3: Segmentation of the left  atrial  appendage from 3D images

Views of the Left Atrial Appendage

• Variable shapes• 1 to 19 cm3

• Function ?• Has to be ablated sometimes

Page 4: Segmentation of the left  atrial  appendage from 3D images

Motivation

Current implementation:• LAA is represented as a short trunk in the model• Current framework not flexible enoughto grow into highly variable shape

Page 5: Segmentation of the left  atrial  appendage from 3D images

Motivation

Motivation of my master thesis:• Addition of an Automatic SegmentationAlgorithm of the Left Atrial Appendagein the Philips Framework

Page 6: Segmentation of the left  atrial  appendage from 3D images

Plan

I. Current Method at PhilipsII. Actual WorkIII. Results and Future Work

Page 7: Segmentation of the left  atrial  appendage from 3D images

Philips Aachen method

New Image Segmentation Chain Segmented Image

1. HeartDetection

2. Parametric Adaptation(Similarity)

4. Deformable Adaptation

3. Parametric Adaptation

(Piecewise Affine)

Page 8: Segmentation of the left  atrial  appendage from 3D images

Philips Aachen method

New Image Segmentation Chain Segmented Image

1. HeartDetection

2. Parametric Adaptation(Similarity)

4. Deformable Adaptation

3. Parametric Adaptation

(Piecewise Affine)

vadap = T[v]E = Eext[T]

Page 9: Segmentation of the left  atrial  appendage from 3D images

Parametric Adaptation,Deformable Models

• Use External Energy :

Page 10: Segmentation of the left  atrial  appendage from 3D images

Philips Aachen method

New Image Segmentation Chain Segmented Image

1. HeartDetection

2. Parametric Adaptation(Similarity)

4. Deformable Adaptation

3. Parametric Adaptation

(Piecewise Affine)

Free motion for vadap

E = Eext[v]+ α Eint[v]vadap = T[v]E = Eext[T]

Page 11: Segmentation of the left  atrial  appendage from 3D images

Deformable Models

• Internal Energy

Page 12: Segmentation of the left  atrial  appendage from 3D images

Plan

I. Current Method at PhilipsII. Actual Work

1. Segment manually 17 patients LAA2. Modify Philips models

1. Interface Left Atrium - Left Atrium Appendage2. Mesh which inflate

3. Code an automatic mesh-inflation algorithm1. External Energy2. Threshold between LAA – Background3. Internal Energy

III. Results and Future Work

Page 13: Segmentation of the left  atrial  appendage from 3D images

Plan

I. Current Method at PhilipsII. Actual Work

1. Segment manually 17 patients LAA2. Modify Philips models

1. Interface Left Atrium - Left Atrium Appendage2. Mesh which inflate

3. Code an automatic mesh-inflation algorithm1. External Energy2. Threshold between LAA – Background3. Internal Energy

III. Results and Future Work

Page 14: Segmentation of the left  atrial  appendage from 3D images

Plan

I. Current Method at PhilipsII. Actual Work

1. Segment manually 17 patients LAA2. Modify Philips model

1. Interface Left Atrium - Left Atrium Appendage2. Mesh which inflate

3. Code an automatic mesh-inflation algorithm1. External Energy2. Threshold between LAA – Background3. Internal Energy

III. Results and Future Work

Page 15: Segmentation of the left  atrial  appendage from 3D images

Model Modification

Page 16: Segmentation of the left  atrial  appendage from 3D images

Plan

I. Current Method at PhilipsII. Actual Work

1. Segment manually 17 patients LAA2. Modify Philips model

1. Interface Left Atrium - Left Atrium Appendage2. Mesh which inflate

3. Code an automatic mesh-inflation algorithm1. External Energy2. Threshold between LAA – Background3. Internal Energy

III. Results and Future Work

Page 17: Segmentation of the left  atrial  appendage from 3D images

External and Internal Energies

Edge-based

Region-based

MeshReference

TriangleRegularization

Curvature N-GonRegularization

External Energy Internal Energy

Page 18: Segmentation of the left  atrial  appendage from 3D images

External Energy :Edge-Based

• No specific features

Page 19: Segmentation of the left  atrial  appendage from 3D images

External Energy :Region-Based

•Gray Value Above or Under ?

Page 20: Segmentation of the left  atrial  appendage from 3D images

External Energy :Region-Growing

•Gray Value Still Above (Under) ?•Already Annotated ?

Page 21: Segmentation of the left  atrial  appendage from 3D images

External Energy :Region-Growing

Gray Value Still Above (Under) ?Already Annotated ?•Gray Value Still Above (Under) ?•Already Annotated ?

Page 22: Segmentation of the left  atrial  appendage from 3D images

External Energy :Region-Growing

Gray Value Still Above (Under) ?Already Annotated ?•Gray Value Still Above (Under) ?•Already Annotated ?

Page 23: Segmentation of the left  atrial  appendage from 3D images

Threshold LAA-Myocardium

Page 24: Segmentation of the left  atrial  appendage from 3D images

Threshold LAA-Myocardium

Page 25: Segmentation of the left  atrial  appendage from 3D images

Threshold LAA-Myocardium

oMinimization of classification erroro Stop when Area1 = Area2

Page 26: Segmentation of the left  atrial  appendage from 3D images

Internal Energy :Mesh Reference

• Updated Mesh

Page 27: Segmentation of the left  atrial  appendage from 3D images

Internal Energy : Triangle Regularization

• Approximate each triangle by a rotated and scaled equilateral triangle

Page 28: Segmentation of the left  atrial  appendage from 3D images

Internal Energy :Curvature

• Remove the peaks

Page 29: Segmentation of the left  atrial  appendage from 3D images

Internal Energy :Curvature

• Remove the peaks

Page 30: Segmentation of the left  atrial  appendage from 3D images

Internal Energy :N-Gon Regularization

• Approximate each “N-Gon” by a rotated and scaled regular N-Gon

Page 31: Segmentation of the left  atrial  appendage from 3D images

External and Internal Energies

Edge-based

Region-based

MeshReference

TriangleRegularization

Curvature N-GonRegularization

External Energy Internal Energy

Page 32: Segmentation of the left  atrial  appendage from 3D images

Plan

I. Current Method at PhilipsII. Actual WorkIII. Results and Future Work

Page 33: Segmentation of the left  atrial  appendage from 3D images
Page 34: Segmentation of the left  atrial  appendage from 3D images

Results

Page 35: Segmentation of the left  atrial  appendage from 3D images

Results

• Main problem : loops -> repair

Page 36: Segmentation of the left  atrial  appendage from 3D images

Results

Specificity = True Pos. /(True Pos. + False Neg.)

Quality = True Pos. /(True Pos. + False Pos.)

Page 37: Segmentation of the left  atrial  appendage from 3D images

Results

0

20

40

60

80

100

Left Atrial Appendage Inflation ResultsSpecificity = True Pos. / (True Pos. + False Neg.) Quality = True Pos. / (True Pos. + False Pos.)

Sum up: • almost all segmented voxels really belong to LAA• but the mesh doesn’t inflate enough

Page 38: Segmentation of the left  atrial  appendage from 3D images

Results

(1) (10)

(5)(7)

Page 39: Segmentation of the left  atrial  appendage from 3D images

Majors Fails

(11) (14)

Page 40: Segmentation of the left  atrial  appendage from 3D images

Possible future works

• Improve the loop repair :– Freeze vertices– Better correction

• Find a new internal energy ?

Page 41: Segmentation of the left  atrial  appendage from 3D images

Thank you for your attention !

Any Questions ?