Interactive Navigation and Bronchial Tube Tracking in Virtual Bronchoscopy

1
Interactive Navigation and Bronchial Tube
Tracking in Virtual Bronchoscopy
Contact Dr. Pheng-Ann Heng Dept. of Computer
Science Engineering The Chinese University of
Hong Kong Shatin, HONG KONG

• Email pheng_at_cse.cuhk.edu.hk
• Homepage http//www.cse.cuhk.edu.hk/pheng/

2
Aim

• Simulation of bronchoscopy in an interactive
  virtual environment
• Non-invasive lung diagnosis and surgical
  bronchoscopy planning
• Virtual bronchial tree tracking, navigation and
  visualization

3
System Layout
4
Bronchial Tree Navigator

• 3D pen versus bronchoscope

5
Our Bronchoscope

• Clinical Bronchoscope
• Have certain thickness
• Cannot pass stenosis
• May embed surgical tools
• Biopsy, injection,
• Minimally invasive
• Generally safe but may cause anxiety and coughing
• In/out and rotate only

• 3D Pen Virtual Scope
• Virtually no dimension
• Able to navigate freely
• For diagnosis only
• Cannot perform surgical operation
• Totally non-invasive
• No pain and safe
• Off-line operation
• Flexible control
• Six degree of freedom

6
Virtual Views

• Airway view
• Maximal Intensity Projection (MIP) view
• Tri-plane cutting view

7
Bronchial Tube Tracking

• Manual tracking
• Tedious and labor-intensive
• Fully automatic tracking
• May not be reliable, especially when dealing with
  ambiguities
• Our semi-automatic tracking technique
• Intelligent Path Tracker
• Provides automation and enough user control

8
Intelligent Path Tracker

• Extension of 2D image Intelligent Scissor by
  Mortensen and Barrett to 3D volume
• Consider each voxel as a node and connect all
  neighbouring voxels by edges
• Assign weight to each edge according to
• Laplacian zero-crossing,
• Gradient magnitude, and
• Gradient direction

9
Vessel Tracking

• Apply Dijkstras algorithm to find an optimal
  branching structure from a root
• Improved by node-pruning
• User control
• Root of the bronchial tree and the interested
  terminals
• Under a 3D virtual environment - Virtual Workbench

10
Segmentation Feature

• Accurate
• Human control avoids blind segmentation without
  considering trivial or ambiguity cases
• Reliable
• Sub-voxel dimension broken vessels are
  automatically repaired and accurately traced
• Efficient
• On a SGI Octane R10000 MXI, user can locate the
  branches of a lung volume interactively by simple
  clicking

11
Volume Rendering Engine

• Project a 3D volume onto an image plane
• Simulate Translucency
• Computation Intensive
• Accelerations
• IsoRegion Leaping
• Distributed Rendering

12
IsoRegion Leaping

• Exploit coherence inside the volume
• IsoRegion
• Cubic homogeneous voxel block
• We store the dimension of all IsoRegions

13
Leaping Algorithm

• Pre-compute the composed color and translucency
  of ray segments of all possible lengths
  containing samples of same voxel value
• The composition process is thus accelerated by
  parts, going through IsoRegions

14
Distributed Rendering

• Distribute the rendering process on a cluster of
  heterogeneous workstations
• Communicate through TCP/IP
• Image-based task partitioning
• Client-server model with load balancing
• Deliver over 10 frames per second for 2563-gt2562
  projection

15
Conclusion

• An interactive virtual environment for simulation
  of bronchoscopy is developed
• The system makes use of bronchial tube tracking
  to provide navigation information, and high
  performance volume rendering engine to achieve
  interactive fly-through
• This is to reduce possible risk and help
  pre-planning clinical bronchoscopy