1. Fully Automatic Blood vessel Branch Labeling Lei Chen Supervisors: Ir. Jan Bruijns Prof. Bart M. ter Haar Romeny

2. Biomedical Engineering PAGE 116-6-2015 Content Pathophysiology background −Brain aneurysm −Coronary artery disease (CAD) Therapy with medical imaging technology −3D rotational angiography (3DRA) Research software environment −The demo program My research questions −Faster blood vessel analysis −Surface wave propagation −Faster branch labeling

3. Background- Pathophysiology Aneurysm 1/15 people in the United States Rupture-bleeding into the brain From www.nhlbi.nih.gov PAGE 216-6-2015

4. Background- Pathophysiology CAD - atherosclerosis Blood clots block blood vessel to cause a heart attack From www.nhlbi.nih.gov PAGE 316-6-2015

5. Medical Imaging Technology Diagnosis of aneurysms and CAD 3D rotational angiography (3DRA) (From www.brainaneurysm.com) PAGE 416-6-2015

6. Therapy with 3DRA Treatment Brain aneurysms Endovascular Coiling Surgical Clipping CAD Place a stent (From www.brainaneurysm.com) PAGE 516-6-2015

7. Software Environment Research tools Linux system Made by C++ Display window Control window −Seven sub-windows with interaction sub-sets Biomedical Engineering PAGE 616-6-2015

8. Biomedical Engineering PAGE 716-6-2015 Research Questions Blood vessel analysis – Shape parameters extraction −Aneurysm volume, Stenosis size, etc. −Original analysis algorithm: −Fully automatic branch labeling of Voxel Vessel Structures My work: improved branch labeling algorithm −Surface wave propagation −Centerlines extraction −Bifurcations detection

9. Research Questions Shape parameters extraction Biomedical Engineering PAGE 816-6-2015 Input datasets Segmentation Branches Labeling Computer-Aided Treatment Planning

10. Original Algorithm Fully automatic branch labeling of voxel vessel structures Biomedical Engineering PAGE 916-6-2015 Detect extremities Extract skeleton Labeling branches Create vessel graphs Create node geometry

11. Original Algorithm: 1 st Step - The Extremities Detection Wave propagation algorithm Double and multiple Zahlten waves Biomedical Engineering PAGE 1016-6-2015 Find seeds Initial wave Double or multiple Waves Direction adjustment Get extremities

12. Wave propagation The volume wave propagation Biomedical Engineering PAGE 1116-6-2015

13. My work: Faster wave propagation-Task 1 Acceleration algorithms Down sampling −Resize the datasets −Drawbacks −Lost information −Decrease accuracy Surface wave propagation −Propagate on the surface of the blood vessel −Seed point selection −Extremities detection Biomedical Engineering PAGE 1216-6-2015

14. Faster Wave Propagation : Down Sampling Acceleration algorithm Down sampling −2*2*2 original voxels -> 1 new voxel −Original volume -> smaller Method: Average as threshold Biomedical Engineering PAGE 1316-6-2015 Processing

15. Down Sampling Result Analysis Down sampling Result: the false and the lost extremity Comparison: − The saving computation time − The number and position of the extremities Biomedical Engineering PAGE 1416-6-2015 Datasets Old / New method False/Lost extremities Improved speed/ratio Extremitieselapsed time 128*128*128 Aneurysm phantom 7 / 50.457 / 0.342s0/20.115s/25% Stenosis122 / 200.337 / 0.281s1/30.06s/17% Stenosis1_po st 54 / 530.512 / 0.461s6/10.05s/10% 256*256*256 Large_ane0259 / 4824.119 / 8.375s12/2315.74s/65% Large_ane50179 / 1805.510 / 4.487s27/821.02s/19 % Large_ane51177 / 9333.770 / 14.969s18/10318.8s/56%

16. Faster Wave Propagation : Surface Wave Propagation Biomedical Engineering PAGE 1516-6-2015 Seed point selection −Detect seed points on the boundary layers −Initialize the 1 st wave Set special label to the surface voxel Adjust the extremities selection method −Keep the waves continuity −Detect the false extremities −A center voxel in the stop wave – an extremity

17. Wave propagation on surface Biomedical Engineering PAGE 1616-6-2015 The surface wave propagation

18. Result of Surface Wave Propagation 26 – connected path between the original and new extremities Biomedical Engineering PAGE 1716-6-2015

19. Acceleration of Branch Labeling – Task 2 Extract the center lines and the bifurcations Problem: bifurcation correction - thinning method To combine the first two original steps to one Biomedical Engineering PAGE 1816-6-2015

20. Improved Branch Labeling Algorithm The wave moving −Simulation by a probe −Single wave normal calculation Methods −Propagate from the root to the branches −Propagate from the branches to the root Biomedical Engineering PAGE 1916-6-2015

21. Bifurcation Detection Centerlines extraction Single wave normal calculation −A plane F (x, y, z) = ax + by + cz + d = 0 −A normal vector (a, b, c) −The voxel points in a wave put into a linear equations: AX=0 Biomedical Engineering PAGE 2016-6-2015

22. Single wave normal calculation homogeneous least square problem Singular value decomposition Matrix : A=USV* A: consist of all the voxels U: m-by-m unitary matrix S: m-by-n diagonal matrix with nonnegative Eigen values V: n-by-n unitary matrix include the Eigen vectors Biomedical Engineering PAGE 2116-6-2015

23. Single wave normal calculation Single wave normal result Biomedical Engineering PAGE 2216-6-2015

24. Center points and Bifurcation Detection Method 1.Store the wave parameters, the average diameter etc. 2.Find and label the abnormal wave 3.Detect the disconnected waves 4.Find the bifurcation −Go back to the abnormal wave to label the bifurcation 5.Store the bifurcation voxels Biomedical Engineering PAGE 2316-6-2015 Move waves Store wave parameters Label (abnormal) waves/branch Label bifurcation Position bifurcation point

25. Bifurcation Detection Result Propagate from the root to the branch Biomedical Engineering PAGE 2416-6-2015

26. Bifurcation Detection Validation Compare to the original algorithm Biomedical Engineering PAGE 2516-6-2015

27. Surface Wave Propagation - Validation The validation and comparison between the original and new wave propagation −The saving computation time −The number and position of the extremities −Decidability of the old and new extremity on the same tip: 26 – connected path Biomedical Engineering PAGE 2616-6-2015 Datasets Old / New method Lost/False extremities Improve to ratio ExtremitiesElapsed time 128*128*128 Aneurysm p7 / 60.663 / 0.37s1/044% Stenosis122 / 220.328 / 0.18s0/045% Stenosis1_post51 / 490.297 / 0.19s3/136% 256*256*256 Large_ane0258 / 5348.543 / 15.32s12/1868% Large_ane50155 / 1207.229 / 4.76s46/1134 % Large_ane5281 / 6836.39 / 11.3118/569%

28. Surface Wave Propagation Validation Compare to the original algorithm Biomedical Engineering PAGE 2716-6-2015

29. Bifurcation Detection Validation Compare to the original algorithm Biomedical Engineering PAGE 2816-6-2015

30. Conclusion Biomedical Engineering PAGE 2916-6-2015 Surface wave propagation gives correct extremity detection results. the detection time is decreased at 34% to 69%. The centerlines and bifurcations extraction gives correct and visually acceptable results.

31. Future Work Biomedical Engineering PAGE 3016-6-2015 Detect the bifurcation The bifurcation adjustment Centerlines validation

32. Questions? Biomedical Engineering PAGE 3116-6-2015 Thanks for your attention!