Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE...
Transcript of Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE...
![Page 1: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/1.jpg)
Computational Anatomy: Multi-organ Modeling and Analysis in
Abdominal CT
Marius George Linguraru, D.Phil.
GLOBAL HEALTH – 22nd October 2012
![Page 2: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/2.jpg)
Site Map
Introduction
Established Segmentation
Priors in Medical Image Data
Segmentation and Simulation
![Page 3: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/3.jpg)
Utilization rates of CT (♦); nuclear medicine (▪); and MRI (▴) in Medicare fee-for-service population, 1998–2008.
[Levin D C et al. AJR 2011]
![Page 4: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/4.jpg)
4
Courtesy of Reuben Mezrich MD, Ph.D.
![Page 5: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/5.jpg)
5
Courtesy of Reuben Mezrich MD, Ph.D.
![Page 6: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/6.jpg)
6
Courtesy of Reuben Mezrich MD, Ph.D.
![Page 7: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/7.jpg)
7
Courtesy of Reuben Mezrich MD, Ph.D.
![Page 8: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/8.jpg)
8
![Page 9: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/9.jpg)
The human observer may be the greatest source of
variability in the image interpretation chain
![Page 10: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/10.jpg)
Courtesy of Nabile, Safdar, MD
![Page 11: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/11.jpg)
Courtesy of Nabile, Safdar, MD
![Page 12: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/12.jpg)
Vision, Light, Luminance, Motion
Courtesy of Nabile, Safdar, MD
![Page 13: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/13.jpg)
Vision, Light, Luminance, Motion
Courtesy of Nabile, Safdar, MD
![Page 14: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/14.jpg)
Mammography
tumor
![Page 15: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/15.jpg)
![Page 16: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/16.jpg)
Clinical Challenges of Segmentation
Large variations on organ shape, size, location.
Similar appearance.
Unusual/abnormal anatomy.
Fast motion.
Use anatomical and physiological constraints typical tomedical image data.
In clinical practice - manual measurements (often 2D)– high intra- and inter-operator variability.
– time consuming – expensive.
Loads of data!
Need: quantitative, robust, accurate, repeatable.
![Page 17: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/17.jpg)
Computer-Assisted Radiology
CAD applications focusedon organ- or disease-based applications.
Radiologists analyze the entire image data.– Organ-by-organ.
– Slice-by-slice.
Migration toward the automated simultaneous analysisof multiple organs for comprehensive diagnosis.
PancreasLiver
Right
Kidney
Left
Kidney
Spleen
![Page 18: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/18.jpg)
Clinical Protocol
Diagnostic
– Contrast enhanced CT – 3 Phases
Serial Monitoring
– Manual measurements
– Limitations
Pre-Contrast Arterial Phase Venous Phase
![Page 19: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/19.jpg)
Site Map
Introduction
Established Segmentation
Priors in Medical Image Data
Segmentation and Simulation
![Page 20: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/20.jpg)
Segmentation Techniques
Lower level– Pixel-based
– Intensity, gradients.
– Region-based
Thresholding.
Edge detection.
Histogram-based.
Mathematical morphology.
Region growing/clustering.
Cannot handle variability![espin086.wordpress.com]
[Linguraru et al., Med Imag Anal 2012]
![Page 21: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/21.jpg)
Higher Level Segmentation
Partial Differential Equations– Snakes
[Kass and Terzopoulos, IJCV 1987]
– Splines
– Deformable models
– Level sets
[Osher and Sethian, J Comput Phys 1988]
Need initialization.
Computationally (in)efficient.
Parametric.
Handle topological changes.
http://www.tnt.uni-hannover.de
http://www.mathworks.com
![Page 22: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/22.jpg)
Higher Level Segmentation
Graph- based Partitioning– Min-cut (graph-cut)
[Wu and Leahy, IEEE TPAMI 1993]
– Random walker
[Grady, IEEE TPAMI 2006]
Need initialization.
Computationally efficient.
Globally optimal.
Any topology.
Multiple objects.
[Lai et al., Comp Aid Geom Design 2009]
[Linguraru et al., Med Imag Anal 2012]
![Page 23: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/23.jpg)
Higher Level Segmentation
Model-based– Atlas-based
– Active Shape Models
– Active Appearance Models[Cootes and Taylor, BMVC 2006]
Need point correspondences.
Sensitive to training set.
Match to a new topology.
Multiple objects.
Hybrids![Ionita and Cootes. IEEE ICCV Workshop 2011]
[Linguraru et al., Med Phys 2010]
![Page 24: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/24.jpg)
Site Map
Introduction
Established Segmentation
Priors in Medical Image Data
Segmentation and Simulation
![Page 25: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/25.jpg)
Visible Human Project (NLM)
Image library of volumetric data representing complete, normal adult male and female anatomy.
MRI/CT/anatomical images.
Models of the body.
Insight Toolkit (ITK).
Columbia University found several errors in anatomy textbooks.
![Page 26: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/26.jpg)
Anatomical Analysis
Organ size is an indicator of disorders.
Shape is locally variable in organs – globalconstraints.
Soft tissue enhancement helps detecting abnormality.
Organ geometry and enhancement are 3D.
![Page 27: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/27.jpg)
Priors in Medical Data
Location
Shape
Appearance
Interaction
Training data.
Integration.
![Page 28: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/28.jpg)
Probabilistic AtlasOrgan positions normalized to anatomical landmarks.
Linear transformation: translation, rotation.
Probabilities of liver in the abdominal cavity.
…
( )( )∑∈
−=Pp
plocation OpSAE |ln)(
![Page 29: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/29.jpg)
Shape Distribution
Linear transformation: translation, rotation, scaling.Preserves shape.
Statistical Shape Models – from a population.
[Linguraru et al., MICCAI 2010]
s1
s2
( ) ( )( ) 12
21 =− sHsH
( ) ( )( ) 02
21 =− sHsH
[Okada et al. , MICCAI 2008]
( ) ( ) ( )( ) ( ) ( )∫∫ −= dxsHdxsHsHsHssD 11
2
2121 ,
Dissimilarity Metric
![Page 30: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/30.jpg)
Intensity Model
Organ NCP
PVP
BackgroundNCP
PVP
+−=
)|()|()|()|(
)|()|(ln)(
BIPBIPOIPOIP
OIPOIPOR
p
pvppvp
p
ncpncp
p
pvppvp
p
ncpncp
p
pvppvp
p
ncpncp
p
![Page 31: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/31.jpg)
Enhancement Model
NCP PVP
Slow enhancement
(high penalty)
Rapid enhancement
(low penalty)
( )pvpncp
p
ncp
p
pvp
p
IIE
σσ2
2−
=
![Page 32: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/32.jpg)
Model Integration - Energy
Appearance
Location
Shape
( ) ( ) ( ) ( ) ( )AEAEAEAEAE shapelocationenhanceensity +++= int
Graph
![Page 33: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/33.jpg)
Graph Cuts
1. Image can be decomposed into a graph of nodes and edges.
2. Background (B) and Object (O) seeds initialize a segmentation.
3. Node are connected to terminals and are inter-connected.
4. Node connections have costs.
5. A cut corresponds to the minimum cost/maximum flow of the total segmentation energy.
( ) ( ) ( )AEAEAE boundaryregion +=
[Boykov and Jolly: ICCV 2001]
![Page 34: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/34.jpg)
Graph Cuts
Object Terminal
t-link
n-link
Background Terminal
cut
![Page 35: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/35.jpg)
Multi-objects – Multi-phase
[Linguraru et al., Med Imag Anal 2012]
![Page 36: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/36.jpg)
Integration - 4D Graph
Intensity Model
4D Graph4D Graph
4D Convolution
4D Filter
Preprocessing
Smooth Register
Seeds
Location Model
Probabilistic Atlas
PatientHistograms
Shape Model
Parzen Density
NCP
PVP
Multi-PhaseCT
[Linguraru et al., Medical Image Analysis 2012]
![Page 37: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/37.jpg)
Integration – 4D Graph( ) ( ) ( ) ( ) ( ) ( )( )∑ =
++++=4
1i shapeboundarylocationenhancedata AEAEAEAEAEAE
( )( )∑∈
−=Pp
plocation OpSAE |ln)(
( ) ( )( ) ( )( ) ( )
==
== →→otherwiseqpdistsPSsPS
sPSsorPSAifAvv
qp
qp
qppqqp
,,/,max
,0}{}{
( ) ( )( ) 1 ELSE 1 THEN , IF }{}{ ==> →→ qppq
qpvvsPSsPS
( ) ∑∑∈
→
∈
→ −+=pp Nqp
pq
Nqp
qpshape vvAE},{
}{
},{
}{ 1)( δδ
( )∑∑∈∈
−+=Bp
p
Op
pdata BRORAE )(1)()( λλ
( )∑∈
+=Pp
penhance EAE211)(
( )pvpncp
p
ncp
p
pvp
p
IIE
σσ2
2−
=
( ) 1 ELSE ,1 THEN ,)( OR )( IF }{}{ ==>−>− →→ qppqncp
q
ncp
p
ncppvp
q
pvp
p
pvp wwIIII σσ
( ) ∑∑∈
→
∈
→ −+=pp Nqp
pq
Nqp
qpboundary wwAE},{
}{
},{
}{ 1)( µµ
−⋅−−
=
== →→ otherwiseqpdist
IIII
AifA
wwInitialize
pvpncp
q
pvp
p
pvp
q
ncp
p
ncp
qp
pqqp ,),(
1
2exp
,0
}{}{
σσ
( ) ( ) ( )( ) ( ) ( )∫∫ −= dxsHdxsHsHsHssD 11
2
2121 ,
![Page 38: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/38.jpg)
Results
![Page 39: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/39.jpg)
Results
![Page 40: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/40.jpg)
Some Organs are More Challenging!
PancreasLiver
Right
Kidney
Left
Kidney
Spleen
![Page 41: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/41.jpg)
P(Liver) P(Spleen)
P(L-Kidney|Spleen)
P(Pancreas|Liver,Spleen)P(R-Kidney|Liver)
P(Gallbladder|Liver )
Hierarchical Inter-Patient Anatomical Variability
Stable organs
1.0
0.18
1.0 0.93
0.68
[Okada et al. , MICCAI Abdominal 2011]
![Page 42: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/42.jpg)
Prediction-based Probabilistic Atlas
P(Pancreas)
P(R-Kidney)
Conventional
P(Gallbladder)
P(Pancreas|Liver,Spleen)
P(Gallbladder|Liver )
P(R-Kidney|Liver)
Hierarchical
![Page 43: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/43.jpg)
Anatomical constraints
Important in surgical planning and guidance.
Abdominal Vessels
Courtesy of Yoshinobu Sato, PhD
![Page 44: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/44.jpg)
Vessel Models
Courtesy of Yoshinobu Sato, PhD
![Page 45: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/45.jpg)
Flow-in-region atlasExtracted organ & aortaOriginal CT
Vessel Models
[Suzuki et al., MICCAI CLIP 2002]
Courtesy of Yoshinobu Sato, PhD
![Page 46: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/46.jpg)
Site Map
Introduction
Established Segmentation
Priors in Medical Image Data
Segmentation and Simulation
![Page 47: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/47.jpg)
Proximity of tumors to intrahepatic veins - patient’s
suitability for surgery/intervention.
Minimally invasive therapies – minimize healthy
tissue damage.
Living donor liver transplant – segmental anatomy.
Segmentation to Intervention
[Madoff DC, et al 2002]
![Page 48: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/48.jpg)
[Pamulapati et al., MICCAI Abdominal 2011]
Segmental Anatomy
![Page 49: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/49.jpg)
Vein Clamping
[Drechsler et al., MICCAI Abdominal 2011]
Simulate effect of vein clamping
– Training
– Planning
– Safety margins
![Page 50: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/50.jpg)
Simulate Catheterization
Localized root and leaf nodes are shown below.
Courtesy of Yoshinobu Sato, PhD
![Page 51: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/51.jpg)
Shortest path findings are performed from all nodes
Simulate Catheterization
Courtesy of Yoshinobu Sato, PhD
![Page 52: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/52.jpg)
Shortest path findings are performed from all nodes
Simulate Catheterization
Courtesy of Yoshinobu Sato, PhD
![Page 53: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/53.jpg)
ConsiderSpeed – motion modeling
(US 25 frames/s + heart 80 b/min)
Size – for pediatrics
Interactive segmentation
– more accurate/preferable
Machine learning
- learn from large data
Human body is well studied
(multiple organs)
[Harvard University]
![Page 54: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/54.jpg)
Acknowledgements
National Institutes of Health, USA – Ronald Summers, MD PhD– Bradford J. Wood, MD
Children’s National Medical Center, USA– Nabile Safdar, MD
Osaka University, Japan– Yoshinobu Sato, PhD – Toshiyuki Okada, PhD
Fraunhofen Institute, Germany– Klaus Drechsler, PhD
Harvard University, USA– Robert Howe, PhD
![Page 55: Global Health 2012 Linguraru · [Wu and Leahy, IEEE TPAMI 1993] – Random walker [Grady, IEEE TPAMI 2006] Need initialization. Computationally efficient. Globally optimal. Any topology.](https://reader036.fdocuments.us/reader036/viewer/2022070111/604c93a54c5ffa66e84ab8e5/html5/thumbnails/55.jpg)
Thank you!
Marius George Linguraru, D.Phil.
GLOBAL HEALTH – 22nd October 2012