Analysis of Tubular Structures

Question 1

Vessel analysis chain

Answer 1

• Image Aquisition
• Preprocessing
• Segmentation
• Grqph Analysis
• Visualization
• Measurement

Question 2

Vessel Image Aquisition

Answer 2

• Imaging with contrast agent
• X-ray, CT, MRA
• Using blood flow properties (top-voxel intensity corresponds to blood velocity)

Question 3

Preprocessing - Vesselness

Answer 3

• distinguish between tubular, blob-like and plate-like Patterns
• Analysis of eigenvalues of hessian matrix

Question 4

Tube Detection

Answer 4

• Hough Transform to detect circles

Question 5

Tube Detection - Region Growing

Answer 5

• simple region growing and region growing with threshold control
• local threshold control (prevent leakages)
• shape control

Question 6

Fuzzy connectedness

Answer 6

• segmentation modeled as minimum cost path problem (Dijkstra)

Question 7

Vessel centerline detection

Answer 7

• compute cost image in which each intensity corresponds to the probability of belonging to a vessel
• determine seed points in vessel
• connect the points via shortest path

Question 8

shortest path methods

Answer 8

• greedy dijkstra: edge based, has to pass through voxel centers
• continuous approach: fast marching: high computation effort, parameterization difficult, more accurate solution can be calculated

Question 9

Template matching

Answer 9

detect centerline by moving template through local image

Question 10

Template matching

Answer 10

detect centerline by moving template through local image

Question 11

selecting the segmentation algorithm

Answer 11

• region growing?
• Watersheds or Fuzzy connectedness?
• retry both after vessel enhancement filter
• shortest path algorithm?
• vessel tracking?
• PhD?
• draw by hand

Question 12

Segmentation postprocessing

Answer 12

• clean up segmentation through opening and closing

- skeletonization for shape analysis

Question 13

Visualization of Vessel Course

Answer 13

• (projected CPR) projection in vessel bounding box determined by centerline
• (Streched CPR) Rotation of plane normal vector into the image def by vector of interest
• (Straightend CPR) Stretching of vessel through stacking of cross-sections

Question 14

Quantification

Answer 14

• Qualitive information: tree depth, branching patterns, relation to surrounding structures
• Quantitative information: vessel diameter, vessel segment length