Registration using Dynamic Data

Transcription

1 Registration using Dynamic Data -- Data acquisition and analysis of dynamic data journee IRMC de Strasbourg Hyewon Seo - CNRS June15th, Univ de Strasbourg

2 Outline Introduction Previous work Overview of the planned work Methodology: Data acquisition and analysis of dynamic data Results Conclusion

3 Introduction Previous work Overview Methodology Results Conclusion Registration? How to: Optimally align two shapes in arbitrary configurations? Central problem in: Image processing Shape acquisition Modeling Used to: Compare Integrate data different measurement devices, viewpoints, times of measurement, different subjects

4 Introduction Previous work Overview Methodology Results Conclusion Registration? Extrinsic vs. intrinsic 2D vs. 3D Rigid vs. non-rigid Inter-subject vs. intra-subject Image vs. surface, boundary vs. volume

5 Introduction Previous work Overview Methodology Results Conclusion Example: 3D intra-subject surface registration Body Modeler Possible shape space Statistical tendencies

6 Introduction Previous work Overview Methodology Results Conclusion Example: 3D intra-subject surface registration Correspondence finding Registration of template model Template model Skeleton + skin mesh attached

7 Introduction Previous work Overview Methodology Results Conclusion Example: 3D intra-subject surface registration Feature based approach to measure the fitting accuracy to guide the conformation 2 steps Joint transformation Vertex displacement +

8 Introduction Previous work Overview Methodology Results Conclusion Registration of the template: problem formulation Joint transformation: Find x such that E(x) is minimized, where x = ([t x1,t y1,t z1,θ x1,θ y1,θ z1,s x1,s y1,s z1 ],θ x2,, s xn ), translation, rotation, and scale of root joint E(x) = Σ P i (x) P i. Distance between corresponding feature points Vertex displacement: Find vertex displacement d that minimizes E(d): E(d) = αe p (d)+βe s (d)+γe d (d). Feature point distance Distortion energy Surface distance

9 Introduction Previous Previouswork work Overview Methodology Results Conclusion Registration of the template: results

10 Introduction Previous work Overview Methodology Results Conclusion Summary of registration in medical imaging Landmark based Anatomical geometrical Segmentation based Rigid Deformable model Voxel property based Reduction to scalars/vectors (moments, principal axes) Using full image content Previous methods do not consider dynamic, time-varying features, despite their increasing clinical relevance!!

11 Introduction Previous work Overview Methodology Results Conclusion Overview 0. Motion data acquisition Movement data in temporal correspondence 1. Analysis of dynamic data Extraction of dynamic features - Compact Tracked mathematical - Similarity motion of densely sampled Kinematic Posterior representation measures probability material of methods: of dynamic between the points, population strain current vector without data analysis shape fields, and its tensors motional behavior inter-frame - Incorporates registration - Segmentation 2 distinct according variations (due to to dynamic anatomical features identity and due to movement) 2. Registration using dynamic data Correspondence computation among dynamic features Transformation 4. Revision of registration Statistical atlas based registration 3. Statistical atlas construction Dimension reduction Joint probabilitstic map

12 Introduction Previous work Overview Methodology Results Conclusion Motion data acquisition 3D scanner With color markers Optical motion capture device High frequency Tagged MRI Temporal correspondence

13 Introduction Previous work Overview Methodology Results Conclusion Analysis of deforming surface e III e III e II e II e I' e I ε I =. e ' I e I e I. e IV e IV e VIII e VIII. e V e VI e V e VI e VII e VII ε VIII = e ' VIII e VIII e VIII Adopt virtual strain gages and perform local strain analysis!!

14 Introduction Previous work Overview Methodology Results Conclusion Strain gage Used to measure the strain of an object Linear strain gage Strain along one direction Rosette strain gage To determine the three independent components of plane strain, three linearly independent strain measures are needed X Y X Y ε a = + cos 2α + γ XY sin 2α X Y X Y ε b = + β + γ XY β X Y X Y ε c = + cos 2γ + γ XY sin 2γ ε + ε ε ε 2 2 ε + ε ε ε 2 2 ε + ε ε ε 2 2 cos 2 sin 2

15 Introduction Previous work Overview Methodology Results Conclusion Analysis of deforming surface Local strain analysis on deforming mesh surface e III e IV e V e II Y e VI Gage C β α e I Gage B Gage A e VII e VIII X ε ε + ε ε ε γ X Y X Y XY A = + cos(2 0) + sin(2 0) ε + ε ε ε γ ε α α X Y X Y XY B = + cos(2 ) + sin(2 ) ε + ε ε ε γ ε β β X Y X Y XY C = + cos(2 ) + sin(2 ) Use measured values ε A, ε B, ε C and compute ε x, ε y, γ xy Compute principal strains using Mohr s circle Average all principal strains normal strain shear strain

16 Introduction Previous work Overview Methodology Results Conclusion Mohr`s circle for strain element Y ε Y Y P Y Y(ε Y, γ YX ) ε x γ YX O γ XY ε x X O Ф X P X C ((ε x + ε Y )/2, 0) ε 1 2φ ε 2 R ε x, ε Y ε Y 1 R = (2 γ ) + ( ε ε ) XY x y X(ε x, γ XY ) ε 1 ε x + ε y = 2 R ε 2 ε x + ε y = + 2 R γ XY Φ= 1 tan 1 γ ( XY ) 2 ε X ε Y ε x, ε y, γ XY => compute principal strain values ε 1, ε 2, and directions Φ

17 Introduction Previous work Overview Methodology Results Conclusion Problems Principal directions NOT tangential to the surface!! Solution: virtual strain gage on triangles Y X A B α β C

18 Introduction Previous work Overview Methodology Results Conclusion Results on a deforming mesh Maximum strain directions and values Maximum strain values on bending knees Maximum strain values on raising legs Maximum strain directions on bending knees

19 Introduction Previous work Overview Methodology Results Conclusion Isocurves based on the maximum strain

20 Introduction Previous work Overview Methodology Results Conclusion Potential application Langer s lines After an Autrian anatomist Karl Langer first discovered by a French surgeon Baron Dupuytren Predominant alignment of collagen fibers within the dermis Impotant for surgical operation Incisions made parallel to Langer's lines may heal better and produce less scarring

21 Introduction Previous work Overview Methodology Results Conclusion Conclusion Registration New objectives Registration using dynamic features Overall framework On-going work on Data acquisition Data analysis Local strain analysis adopting virtual strain gage Extracted features: principal strain values and directions Results

22 Introduction Previous work Overview Methodology Results Conclusion Acknowledgment Prof. M. de Mathelin and Dr. P. Germain Introduction and access to tagged MRI data