Soft Plaque Detection and Automatic Vessel Segmentation Georgia - - PowerPoint PPT Presentation

▶
soft plaque detection and automatic vessel segmentation
SMART_READER_LITE
LIVE PREVIEW

Soft Plaque Detection and Automatic Vessel Segmentation Georgia - - PowerPoint PPT Presentation

Feb 05, 2024 34 likes •706 views

Soft Plaque Detection and Automatic Vessel Segmentation Georgia Tech Shawn Lankton - Arthur Stillman - Emory University Paolo Raggi - Emory University Allen Tannenbaum - Georgia Tech and Emory PMMIA, MICCAI Workshop September 20, 2009

slide-1
SLIDE 1

Soft Plaque Detection and Automatic Vessel Segmentation

PMMIA, MICCAI Workshop September 20, 2009

Shawn Lankton -

Arthur Stillman - Paolo Raggi - Allen Tannenbaum -

Georgia Tech

Emory University Emory University Georgia Tech and Emory

slide-2
SLIDE 2

/44

Vessel Analysis

  • Heart disease
  • Diagnosis and risk
  • Plaque detection

2

slide-3
SLIDE 3

/44

CTA Imagery

  • In-vivo, 3-D scan
  • X-ray attenuation
  • Contrast agent

3

slide-4
SLIDE 4

/44

Soft Plaque Detection

  • Coronary plaques
  • Dangerous
  • Hard to see
  • No good tools

4

slide-5
SLIDE 5

/44

Soft Plaque Detection

  • Coronary plaques
  • Dangerous
  • Hard to see
  • No good tools

4

slide-6
SLIDE 6

/44

Prior Work

  • Saur et. al. “... detection of calcified

coronary plaques...” MICCAI. 2008

  • Brunner et. al. “… classification of calcified

arterial lesions.” MICCAI 2008

  • Renard and Yang: “… detection of soft

plaques …” {SSIAI,ISBI,ICIP} 2008

5

slide-7
SLIDE 7

/44

Objective

  • From a simple initialization…
  • Segment the vessel…
  • Detect all soft plaques.

6

slide-8
SLIDE 8

/44

Some Definitions

7

slide-9
SLIDE 9

/44

Active Surfaces

8

  • Level set implementation
  • Iterative optimization
  • Simple, flexible, principled
slide-10
SLIDE 10

/44

Definitions

  • 9
  • Sethian. Level Set Methods and Fast Marching Methods. 1999

Such that Γ = {x ∈ Ω|φ(x) = 0} An Image I : RN → R on the domain Ω A Surface Γ embedded in φ : RN → R

slide-11
SLIDE 11

/44

Definitions

10

Hφ =    1 φ < −ǫ φ > ǫ smooth

  • therwise

inside

  • utside

δφ =    1 φ = 0 |φ| < ǫ smooth

  • therwise

the surface the rest

slide-12
SLIDE 12

/44

Localized Active Contour Model

11

slide-13
SLIDE 13

/44

Localizing

12

Lankton and Tannenbaum “Localized Region-Based Active Contours,” TIP, 2008

slide-14
SLIDE 14

/44

Localizing

x

12

Lankton and Tannenbaum “Localized Region-Based Active Contours,” TIP, 2008

slide-15
SLIDE 15

/44

Localizing

x

B(x, y)

12

Lankton and Tannenbaum “Localized Region-Based Active Contours,” TIP, 2008

slide-16
SLIDE 16

/44

Localizing

x

B(x, y) B(x, y) · Hφ(y)

12

Lankton and Tannenbaum “Localized Region-Based Active Contours,” TIP, 2008

slide-17
SLIDE 17

/44

Localizing

x

B(x, y) B(x, y) · Hφ(y)

B(x, y) · (1 − Hφ(y))

12

Lankton and Tannenbaum “Localized Region-Based Active Contours,” TIP, 2008

slide-18
SLIDE 18

/44

Localizing

x

B(x, y) B(x, y) · Hφ(y)

B(x, y) · (1 − Hφ(y))

12

Lankton and Tannenbaum “Localized Region-Based Active Contours,” TIP, 2008

slide-19
SLIDE 19

/44

Localizing

x

B(x, y) B(x, y) · Hφ(y)

B(x, y) · (1 − Hφ(y))

12

Lankton and Tannenbaum “Localized Region-Based Active Contours,” TIP, 2008

≤ r.

slide-20
SLIDE 20

/44

Local Regions in 3-D

13

(a) Example Surface (b) Local Region (c) Local Interior (d) Local Exterior

slide-21
SLIDE 21

/44

Localized Contours

14

E(φ) =

slide-22
SLIDE 22

/44

Localized Contours

14

E(φ) =

  • Ωx

δφ(x) dy dx

every point on the contour

slide-23
SLIDE 23

/44

Localized Contours

14

E(φ) =

  • Ωx

δφ(x)

  • Ωy

B(x, y)dy dy dx

all information in a local ball around x

slide-24
SLIDE 24

/44

Localized Contours

14

E(φ) =

  • Ωx

δφ(x)

  • Ωy

B(x, y) · F(I, φ, x, y) dy dy dx

compute an internal energy F

slide-25
SLIDE 25

/44

Localized Contours

+ λ

  • Ωx

δφ(x)∇φ(x)dx

14

E(φ) =

  • Ωx

δφ(x)

  • Ωy

B(x, y) · F(I, φ, x, y) dy dy dx

slide-26
SLIDE 26

/44

Localized Contours

+ λ

  • Ωx

δφ(x)∇φ(x)dx

14

∂φ ∂t

(x) = δφ(x)

  • Ωy

B(x, y)·∇φ(y)F(I, φ, x, y)dy+λδφ(x) div ∇φ(x) |∇φ(x)|

  • ∇φ(x)

E(φ) =

  • Ωx

δφ(x)

  • Ωy

B(x, y) · F(I, φ, x, y) dy dy dx

slide-27
SLIDE 27

/44

Internal Energies

  • Local Uniform Modeling
  • Local Mean Separation

15

slide-28
SLIDE 28

/44

Localized Means

16

µin( µout( ) = ) =

  • Ωy

y Hφ(y) · I(y)dy

  • ·

y Hφ(y)dy

· (1 − Hφ(y)) · I(y)dy − H · · (1 − Hφ(y))dy

  • Ωy
  • Ωy
  • Ωy
slide-29
SLIDE 29

/44

Localized Means

16

µin( µout( ) = ) =

  • Ωy

y Hφ(y) · I(y)dy

  • ·

y Hφ(y)dy

· (1 − Hφ(y)) · I(y)dy − H · · (1 − Hφ(y))dy

  • Ωy
  • Ωy
  • Ωy

in(x)

=

  • ut(x)

=

y B(x, y) · (1

  • y B(x, y) · (1
  • y B(x, y) · (1
  • y B(x, y) · (1
slide-30
SLIDE 30

/44

Local Vessel Segmentation

17

slide-31
SLIDE 31

/44

Vessel Segmentation

  • Simple initialization
  • No leaks
  • Branch handling
  • No shape information

18

slide-32
SLIDE 32

/44

Local Means

19

slide-33
SLIDE 33

/44

Local Uniform Modeling Energy

20

Assumption: The foreground and background are approximately constant locally.

Fum = Hφ(y)(I(y) − µin(x))2 + (1 − Hφ(y))(I(y) − µout(x))2

slide-34
SLIDE 34

/44

Why Uniform Modeling?

  • Enforce similarity
  • Surface expands quickly
  • Move to capture the “vessel wall”

21

slide-35
SLIDE 35

/44

Energy Minimization

22

dφ(x) dt = δφ(x)

  • ˜

Ωy

B(x, y) · δφ(y) ·

  • I(y) − µin(x)

2 −

  • I(y) − µout(x)

2 dy +λdiv ∇φ(x) |∇φ(x)|

  • |∇φ(x)|
slide-36
SLIDE 36

/44

Energy Minimization

22

dφ(x) dt = δφ(x)

  • ˜

Ωy

B(x, y) · δφ(y) ·

  • I(y) − µin(x)

2 −

  • I(y) − µout(x)

2 dy +λdiv ∇φ(x) |∇φ(x)|

  • |∇φ(x)|

Domain restriction:

˜ Ω =Ω ∩ (I < −600 HU)

slide-37
SLIDE 37

/44

Vessel Segmentation

23

LAD RCA

slide-38
SLIDE 38

/44

Vessel Segmentation

23

LAD RCA

slide-39
SLIDE 39

/44

Parameters

  • Radius
  • Smoothness

24

λ = 0.1max(| dφ

dt |)

r = 5

slide-40
SLIDE 40

/44

Soft Plaque Detection

25

slide-41
SLIDE 41

/44

Soft Plaque Detection

  • Two-front approach
  • Inside moves out
  • Outside moves in

26

slide-42
SLIDE 42

/44

Local Mean Separation Energy

27

Yezzi et al. “A Fully Global Approach to Image Segmentation... ,” JVCIR 2002

Assumption: The foreground and background are different locally.

Fms = −(µin(x) − µout(x))2

slide-43
SLIDE 43

/44

Why Means Separation?

  • Enforces differences
  • Interior won’t grow out
  • Exterior won’t move in

28

slide-44
SLIDE 44

/44

Detection Energy

29

+λdiv ∇φ(x) |∇φ(x)|

  • |∇φ(x)|

dφ(x) dt =

  • ˜

Ωy

B(x, y) · δφ(y)

  • I(y) − µout(x)

2 Aout(x) −

  • I(y) − µin(x)

2 Ain(x)

  • dy

Clever initializations are required

slide-45
SLIDE 45

/44

Initial Surfaces

30

Eshrink(φ) =

  • Ωx

δφ(x)

  • Ωy

(B(x, y) · Hφ(y)) y)) dy dx + λ

  • Ωx

δφ(x)∇φ(x)dx Egrow(φ) = −Hφ(x) dx + λ

  • Ωx

δφ(x)∇φ(x)dx

slide-46
SLIDE 46

/44

Initial Surfaces

30

Eshrink(φ) =

  • Ωx

δφ(x)

  • Ωy

(B(x, y) · Hφ(y)) y)) dy dx + λ

  • Ωx

δφ(x)∇φ(x)dx Egrow(φ) = −Hφ(x) dx + λ

  • Ωx

δφ(x)∇φ(x)dx

slide-47
SLIDE 47

/44

Steps for Detection

  • Segment the Vessel
  • Create Initialization
  • Run Local Mean Separation
  • Check for Differences

31

slide-48
SLIDE 48

/44

3-D Example

32

slide-49
SLIDE 49

/44

3-D Example

32

slide-50
SLIDE 50

/44

3-D Example

32

slide-51
SLIDE 51

/44

3-D Example

32

slide-52
SLIDE 52

/44

3-D Example

32

slide-53
SLIDE 53

/44

Detection Results

33

slide-54
SLIDE 54

/44

Coronary Anatomy

  • Coronaries
  • RCA
  • LAD
  • LCX

34

slide-55
SLIDE 55

/44

Coronary Anatomy

  • Coronaries
  • RCA
  • LAD
  • LCX

34

slide-56
SLIDE 56

/44

Coronary Anatomy

  • Coronaries
  • RCA
  • LAD
  • LCX

34

slide-57
SLIDE 57

/44

Coronary Anatomy

  • Coronaries
  • RCA
  • LAD
  • LCX

34

slide-58
SLIDE 58

/44

2-D Results

35

(a) Initial Surfaces (b) Result of Evolution (c) Expert Marking (d) Detected Plaque

slide-59
SLIDE 59

/44

2-D Results

36

(a) Initial Surfaces (b) Result of Evolution (c) Expert Marking (d) Detected Plaque

slide-60
SLIDE 60

/44

3-D Results (LCX)

37

slide-61
SLIDE 61

/44

3-D Results (LAD)

38

slide-62
SLIDE 62

/44

3-D Results (RCA)

39

slide-63
SLIDE 63

/44

3-D Results (RCA)

40

slide-64
SLIDE 64

/44

Results Summary

41

Table 5.1: Results of soft plaque detection in Figures 5.8 and 5.9. Plaque ID Remodeling Vessel Segment Confirmed Detected #1 negative LAD × #2 positive LAD × × #3 positive LAD × × #4 negative LCX × × #5 positive LCX × × #6 negative RCA × × #7 negative RCA × × #8 positive RCA × ×

slide-65
SLIDE 65

/44

Summary & Conclusions

  • Localized analysis makes sense
  • Vessel segmentations are satisfactory
  • Detection identifies 87.5% of plaques

42

slide-66
SLIDE 66

/44

Continuing Research

  • Coupling contour evolution
  • Acquiring additional data
  • Performing more tests

43

slide-67
SLIDE 67

/44

Thank You.

44