PhD presentation: "Computer Graphics and Vision Techniques for the Study of the Muscular
 Fiber Architecture of the Myocardium".

1. Computer Graphics and Vision Techniques
for the Study of the Muscular
Fiber Architecture of the Myocardium
PhD Thesis
Ferran Poveda
Advisors: Enric Martí
Debora Gil

2. Yearly deaths directly related to heart
conditions in the World*:
9.6 milion by 2015
*Data of the World Health Organization
Motivation
PhD Thesis
Ferran Poveda
!2

3. Yearly deaths directly related to heart
conditions in the World*:
9.6 milion by 2015
11.7 million by 2030
*Data of the World Health Organization
Motivation
PhD Thesis
Ferran Poveda
!3

4. Structure
•
Dilated Cardiomyopathy
•
Cardiac ischemia
!
!
!
!
!
Spatial distribution of myocites (architecture)
plays a crucial part in cardiac function
Form and Function
!4
PhD Thesis
Ferran Poveda

5. Structure
•
Dilated Cardiomyopathy
•
Cardiac ischemia
Rhythm
•
Arrythmogenesis
•
Ventricular dyssyncrony
!
!
Spatial distribution of myocites (architecture)
plays a crucial part in cardiac function
Form and Function
!5
PhD Thesis
Ferran Poveda

6. Structure
•
Dilated Cardiomyopathy
•
Cardiac ischemia
Rhythm
•
Arrythmogenesis
•
Ventricular dyssyncrony
Dynamics
•
Diastolic heart failure
Spatial distribution of myocites (architecture)
plays a crucial part in cardiac function
Form and Function
!6
PhD Thesis
Ferran Poveda

7. Lower
1669
Senac
1749
Kherl
1891
•
No consensus on an architectural model
•
Surgical and Histological exploration
•
Torrent-Guasp
1957
Criticised in its repeatability and subjectivity
Controversy
PhD Thesis
Ferran Poveda
!7

8. Objective measurement of the biological structure
DT-MRI
PhD Thesis
Ferran Poveda
!8

9. Objective measurement of the biological structure
DT-MRI
PhD Thesis
Ferran Poveda
!9

10. Objective measurement of the biological structure
DT-MRI
PhD Thesis
Ferran Poveda
!10

11. Objective measurement of the biological structure
DT-MRI
PhD Thesis
Ferran Poveda
!11

12. Use of tractography for representation of cardiac architecture:
•
Complete and objective:
•
segmented volumes
!
!
Current Challenges
!12
PhD Thesis
Ferran Poveda

13. Use of tractography for representation of cardiac architecture:
•
Complete and objective:
•
segmented volumes
•
partial reconstructions
!
Current Challenges
!13
PhD Thesis
Ferran Poveda

14. Use of tractography for representation of cardiac architecture:
•
Complete and objective:
•
•
•
segmented volumes
partial reconstructions
Simplify structure representation
Current Challenges
!14
PhD Thesis
Ferran Poveda

15. Use of tractography for representation of cardiac architecture:
•
Complete and objective:
•
segmented volumes
•
partial reconstructions
•
Simplify structure representation
•
Validation of reconstructions
Current Challenges
!15
PhD Thesis
Ferran Poveda

16. Goal:
•
Provide cardiologists a simplified, objective and validated
representation of the myocardial architecture.
Contributions:
•
Simplification: Achieve more abstract reconstructions
•
Validation: Measure quality and confidence of fiber tracts
•
Objective: Visualization tools and study of the myocardial architecture
Goal and contributions
!16
PhD Thesis
Ferran Poveda

17. •
Fiber tracking and multi-resolution tractography
•
Validation Framework
•
Visualization and evaluation of a conceptual model of the myocardium
Outline
PhD Thesis
Ferran Poveda
!17

18. •
Reconstruction of discrete measurement of cardiac
architecture. Integration of curves on DT-MRI vector field.
Runge-Kutta integration
•
Initiation
•
Vector orientation
•
Termination
Fiber Tracking
PhD Thesis
Ferran Poveda
!18

19. •
Coherence measure:
LoCoh = max(σeig11, σeig12, σeig13)
Fiber Tracking: Initiation
!19
PhD Thesis
Ferran Poveda

20. •
Coherence measure:
LoCoh = max(σeig11, σeig12, σeig13)
Fiber Tracking: Initiation
!20
PhD Thesis
Ferran Poveda

21. •
Orientation problem:
Fiber Tracking: Orientation
!21
PhD Thesis
Ferran Poveda

22. •
Orientation problem:
Automatically computed anatomical axis
Fiber Tracking: Orientation
!22
PhD Thesis
Ferran Poveda

23. •
Orientation problem:
Automatically computed anatomical axis
Fiber Tracking: Orientation
!23
PhD Thesis
Ferran Poveda

24. •
Termination based solely on RK estimation
•
Noise in thin/unstable structures
•
Not impose structural restrictions
•
Later confidence measures
Fiber tracking: Termination
!24
PhD Thesis
Ferran Poveda

25. •
Multi-resolution pyramid:
Origin
F1
R1
Multi-resolution
F2
R2
FN
PhD Thesis
Ferran Poveda
!25
RN

26. •
Application to DT-MRI/Tractography:
Origin
F1
R1
F2
Multiresolution tractography
!26
R2
FN
PhD Thesis
Ferran Poveda
RN

27. •
Gaussian pyramids:
Isotropic + subsampling
•
Structure Preserving Diffusion operator:
Anisotropic + subsampling
•
Discrete Wavelet Transform:
Haar Wavelet decomposition
Multi-resolution Methods
!27
PhD Thesis
Ferran Poveda

28. Origin
F1
R1
F2
R2
FN
RN
Gaussian Filtering
Gaussian Pyramid
!28
PhD Thesis
Ferran Poveda

29. Origin
•
F1
R1
F2
R2
FN
RN
Use anatomical information to guide filtering and preserve
structure: 1st eigenvector guiding diffusion process
Structure Preserving Diffusion
!29
PhD Thesis
Ferran Poveda

30. Origin
•
F1
R1
F2
R2
FN
RN
Use anatomical information to guide filtering and preserve
structure: 1st eigenvector guiding diffusion process
Input data
Iterative processing
Structure Preserving Diffusion
!30
Filtered data
PhD Thesis
Ferran Poveda

31. Origin
F1
R1
F2
R2
FN
RN
•
We use low frequency bands of the pyramid (less detail)
•
Gaussian can also be seen as an application of the family of Mexican Hat Wavelet Transforms:
•
Specialized in valleys and ridges
•
Rather than discontinuities like Haar DWT
Wavelet Transform
!31
PhD Thesis
Ferran Poveda

32. •
Fiber tracking and multi-resolution tractography
•
Validation Framework
•
Visualization and evaluation of a conceptual model of the myocardium
Outline
PhD Thesis
Ferran Poveda
!32

33. •
All results obtained over the Johns Hopkins University database
•
7 healthy specimens
•
Beagle breed dogs
•
Ex-vivo fixed hearts
•
DT-MRI adquisitions
Database
PhD Thesis
Ferran Poveda
!33

34. Local image analysis
Global geometric analysis
Validation framework
!34
PhD Thesis
Ferran Poveda

35. Local image analysis
Angular
Variation
Global geometric analysis
Structure
Preservation
Validation framework
!35
PhD Thesis
Ferran Poveda

36. Local: Angular
PhD Thesis
Ferran Poveda
!36

37. •
Angular variance in multi-resolution:
Methods
Hearts
Local: Angular
PhD Thesis
Ferran Poveda
!37

38. •
Localization:
Local: Angular
PhD Thesis
Ferran Poveda
!38

39. Coherence Measure
Local: Structure
!39
PhD Thesis
Ferran Poveda

40. Coherence Measure
Segmentation
Local: Structure
!40
PhD Thesis
Ferran Poveda

41. Coherence Measure
Segmentation
Local: Structure
!41
Distribution of classes
PhD Thesis
Ferran Poveda

42. Coherence Measure
•
Segmentation
Distribution of classes
Separability based on Linear Fisher Discriminant
Local: Structure
!42
PhD Thesis
Ferran Poveda

43. Local: Structure
!43
PhD Thesis
Ferran Poveda

44. Local image analysis
Global geometric analysis
Validation framework
!44
PhD Thesis
Ferran Poveda

45. Interest in the study of geometric contextualization
•
Contexts of fibers for the study of geometrical behaviour
•
Spline fitting for arc-parameter matching
Global: Geometric contextualization
!45
PhD Thesis
Ferran Poveda

46. •
3D geometric descriptor for each context
•
•
Curvature, Torsion and Distances
of contexts fibers vs representative
1D metric based on
Mahalanobis distance
Global: Geometric contextualization
!46
PhD Thesis
Ferran Poveda

47. •
Mahalanobis metric for comparison
Contextualization
Original
Original
Global: Geometric contextualization
!47
PhD Thesis
Ferran Poveda

48. •
Mahalanobis metric for comparison
Contextualization
Summarization
Processed
Original
Global: Geometric contextualization
!48
PhD Thesis
Ferran Poveda

49. •
Preservation of geometric contextualization:
!
•
~
Statistical study of variability (ANOVA)
Global: Experimental Setup
!49
PhD Thesis
Ferran Poveda

50. •
Study variability according to different factors:
A. Specimens (quality of capture)
B. Methods (all levels of the pyramids in the different
methodologies)
Group data according to factors and multiple test on each factor mean
Ho: μheart_1 = μheart_2 = · · · = μheart_7
μmethod_1 = μmethod_2 = · · · = μmethod_N
Detect interaction between factors
Global: Experimental Setup
!50
PhD Thesis
Ferran Poveda

51. •
Study variability according to different factors:
A. Specimens (quality of capture)
B. Methods (all levels of the pyramids in the different
methodologies)
Group data according to factors and multiple test on each factor mean
Ho: μheart_1 = μheart_2 = · · · = μheart_7
μmethod_1 = μmethod_2 = · · · = μmethod_N
Detect interaction between factors
Global: Experimental Setup
!51
PhD Thesis
Ferran Poveda

52. Two experiments:
1. Stability across specimens and methods:
Mahalanobis
2. Preservation of geometrical features:
Difference of Mahalanobis
Global: Experimental Setup
!52
PhD Thesis
Ferran Poveda

53. Analysis for complete dataset
2-way ANOVA:
!
Heart Factor: p < 0.000
Method Method: p < 0.000 α = 0.05
Interaction: p = 0.028
1-way ANOVA:
Hearts
Methods
Global: Exp 1
PhD Thesis
Ferran Poveda
!53

54. Analysis for complete dataset
2-way ANOVA:
!
Heart Factor: p < 0.000
Method Method: p < 0.000 α = 0.05
Interaction: p = 0.028
1-way ANOVA:
Hearts
Methods
Global: Exp 1
PhD Thesis
Ferran Poveda
!54

55. 2-way ANOVA:
!
Heart Factor: p < 0.000
Method Method: p < 0.000 α = 0.05
Interaction: p = 0.476
1-way ANOVA:
Hearts
Analysis discarding
outlying methods
Methods
Global: Exp 1
PhD Thesis
Ferran Poveda
!55

56. 2-way ANOVA:
!
Heart Factor: p < 0.000
Method Method: p < 0.000 α = 0.05
Interaction: p = 0.731
1-way ANOVA:
Hearts
Analysis discarding
outlying methods
Methods
Global: Exp 2
PhD Thesis
Ferran Poveda
!56

57. •
Fiber tracking and multi-resolution tractography
•
Validation Framework
•
Visualization and Interpretation
•
Visualization Tools
•
Evaluation of a conceptual model of the myocardium
Outline
PhD Thesis
Ferran Poveda
!57

58. •
Fiber color coding
•
Multi-resolution
•
Reconstruction confidence
Visualization
PhD Thesis
Ferran Poveda
!58

59. •
Global coordinate axes of reference
Visualization: Fiber coloring
!59
PhD Thesis
Ferran Poveda

60. •
Local coordinate axes of reference
Visualization: Fiber coloring
!60
PhD Thesis
Ferran Poveda

61. •
Local coordinate axes of reference
•
Anatomically coherent color mapping
Visualization: Fiber coloring
!61
PhD Thesis
Ferran Poveda

62. Binary color coding
Binary transparency coding
Visualization: Fiber coloring
!62
PhD Thesis
Ferran Poveda

63. Visualization: Multi-resolution
!63
PhD Thesis
Ferran Poveda

64. Multi-resolution visualization
!64
PhD Thesis
Ferran Poveda

65. Visualization: Multi-resolution
!65
PhD Thesis
Ferran Poveda

66. •
Context vs geometric mean:
Geometric mean
Original
Visualization: Confidence
!66
PhD Thesis
Ferran Poveda

67. •
Context vs geometric mean:
Geometric mean
•
Original
Local geometric metric:
Visualization: Confidence
!67
PhD Thesis
Ferran Poveda

68. •
Control streamtube
•
Thickness
•
Color
•
Noise reduction
•
Ease interpretation
Visualization: Confidence
!68
PhD Thesis
Ferran Poveda

69. •
poseso
F. Torrent-Guasp
Helical Ventricular Myocardial Band
Evaluation of a conceptual model: HVMB
!69
PhD Thesis
Ferran Poveda

70. •
Key-points of the Helical Ventricular Myocardial Band:
•
Four segments of the conceptual model
•
Helicoidal connectivity across myocardium
Evaluation of a conceptual model: HVMB
!70
PhD Thesis
Ferran Poveda

71. Full-scale reconstruction
Rubber band model
Right Segment
Left Segment
Descendent Segment
Ascendent Segment
Segments
PhD Thesis
Ferran Poveda
!71

72. A
E
C
B
Lateral superior view
D
Inferior view
Basal Loop
Apical Loop
Segments
PhD Thesis
Ferran Poveda
!72

73. I
A
C
H
D
E
Anterior view
Posterior view
Segments
PhD Thesis
Ferran Poveda
!73

74. Superior view
Segments
PhD Thesis
Ferran Poveda
!74

75. •
a multi-res una sola fibra
Schematic
of the HVB
Band connectivity
!75
PhD Thesis
Ferran Poveda

76. •
a multi-res una sola fibra
Schematic
of the HVB
Band connectivity
!75
PhD Thesis
Ferran Poveda

77. •
a multi-res una sola fibra
Schematic
of the HVB
Band connectivity
!75
PhD Thesis
Ferran Poveda

78. B
D
A
G
•
B
E
C
E
a multi-res una sola fibra
F
D
G
f
C
Schematic
of the HVB
A
Band connectivity
!75
PhD Thesis
Ferran Poveda

79. •
a multi-res una sola fibra
Repeatability
PhD Thesis
Ferran Poveda
!76

80. •
a multi-res una sola fibra
Repeatability
PhD Thesis
Ferran Poveda
!77

81. •
Multi-resolution is a good tool heart architecture interpretation
and representation
•
Wavelet, best to contextualize anatomical information
•
SPD, best to filter data
Conclusions: Simplification
!78
PhD Thesis
Ferran Poveda

82. •
Validation framework comprehensive measures
•
Local measures able to detect worst captures but are not
enough
•
Global geometric features must be taken into account
•
First level of reduction
Conclusions: Validation
!79
PhD Thesis
Ferran Poveda

83. •
Visualization tools allowed showing evidences of the structure of
the HVMB of Torrent-Guasp
•
Band segments
•
Helical continuity
Conclusions:
Visualization and interpretation
!80
PhD Thesis
Ferran Poveda

84. •
Broaden database (species, conditions, specimens)
•
Use of geometrical descriptors
•
•
•
Shapes/Configurations
Applications
Scale-space
Future Work
PhD Thesis
Ferran Poveda
!81

85. •
F. Poveda, D. Gil, E. Martí-Godia, A. Andaluz, M. Ballester, F. Carreras.
Helical Structure of the Cardiac Ventricular Anatomy Assessed by
Diffusion Tensor Magnetic Resonance Imaging Multi-Resolution
Tractography. Revista Española de Cardiología., 66(10):782-790,
2013. + Cover Image - IF 3.204
•
F. Poveda, D. Gil, E. Martí-Godia, M. Ballester, F. Carreras. Helical
structure of ventricular anatomy by diffusion tensor cardiac MR
tractography. Journal of the American College of Cardiology:
Cardiovascular Imaging, 5(7):754-5, 2012. + Cover Image - IF 6.703
•
F. Poveda, D. Gil, E. Martí-Godia. Multi-resolution DT-MRI Cardiac
Tractography. In book Statistical Atlases and Computational Models of
the Heart. Imaging and Modelling Challenges. International
Conference on Medical Image Computing and Computer Assisted
Intervention (MICCAI), 270-277, 2013.
•
F. Poveda, D. Gil, T. Gurguí, E. Martí-Godia. Multi-resolution myocardial
architecture study. In proceedings of Congreso Español de Informática
Gráfica. 165-6, 2012.
Publications
PhD Thesis
Ferran Poveda
!82

86. Computer Graphics and Vision Techniques
for the Study of the Muscular
Fiber Architecture of the Myocardium
PhD Thesis
Ferran Poveda
Advisors: Enric Martí
Debora Gil