RealTime Volume Rendering for Virtual Colonoscopy

1
Real-Time Volume Rendering for Virtual
Colonoscopy
Wei Li, Arie Kaufman, and Kevin Kreeger

• Departments of Computer Science
• State University of New York at Stony
  Brook
• Stony Brook, NY 11794-4400

2
Outline

• Introduction
• The slab structure
• Slab-Based Volume Rendering
• Warping and Caching of Slab Images
• Experimental Results
• Future Work

3
Introduction

• Virtual Colonoscopy

4
Screen shot of the system
5
Introduction

• Challenges
• High quality image
• Large data set
• Interactive Rendering Speed

6
Related Work

• High-end parallel computers
• You 97, Wan 97
• Texture mapping hardware
• 3D texture, 2D texture
• Image-based rendering
• Brady 97, Mueller 99
• Dedicated hardware
• VIZARD II, VolumePro

7
Observations

• Colon is tortuous
• Colon wall is opaque
• 1/5 of voxels are inside the colon
• The camera stays inside the colon

• A small percentage of voxels are visible to a
  given view point.

8
Overview of our approach

• Pre-processing
• Axis-aligned slabs
• Portals
• Rendering
• Visibility culling
• Render or warp slab images

9
Outline

• Introduction
• The slab structure
• Slab-Based Volume Rendering
• Warping and Caching of Slab Images
• Experimental Results
• Future Work

10
The slab structure

• A slab is a sub-volume bounded by an axis-aligned
  box.
• A slab contains as few non-internal voxels as
  possible.

11
Slab Sets of Colon Volume

• Three sets of slabs.
• One set for each major axis

12
Portals
On every shared face of each adjacent slabs, we
define a portal. The portal is the bounding box
of the cross-section of a colon on that shared
face.
13
Construction of the Slabs

• Slice the volume with equidistant parallel planes
  perpendicular to a major axis.
• Find all the connected regions of internal voxels
  between the adjacent planes.
• The bounding box of the connected regions define
  the slabs.
• The process is repeated three times. One for each
  of the three axes.

14
Construction of the Slabs (2)

• Create a binary volume from the original dataset.
  
• Compress the binary volume by merging slices.
• Apply 2D region grow on the slices of the
  compressed volume

15
Compressed Binary Volume
16
Creating the slabs
17
Complexity of the Algorithm

• The algorithm tries to do region grow for all
  (internal) voxels
• Each voxel is inspected only once for whether a
  region grow is necessary.
• Each internal voxel is flagged once
• Time complexity is linear to the number of voxels.

18
Outline

• Introduction
• The slab structure
• Slab-Based Volume Rendering
• Warping and Caching of Slab Images
• Experimental Results
• Future Work

19
Slab-based Volume Rendering

• Determine which set of slabs to use
• Visibility detection
• Obtain the slabs images
• Blending the slab images

20
Visibility Detections

• Project the view frustum to the volume
  coordinates
• Intersect the view frustum with the portals.
• Faster and more accurate than projecting portals
  to the eye coordinates

21
The Blending of Slab Images
22
The Blending of Slab Images (2)
23
Warp the slab images

• Why?
• To reuse the slab images from nearby view points
• To simulate perspective projection with parallel
  hardware

24
Warping based on portal model
25
Warping based on depth image
26
Two-Level Caching of the Slab Images

• Cache in graphics memory
• To save the time for transfer slab images from
  host memory to graphics memory.
• (OpenGL texture object)
• Cache in host memory
• To save the time for volume rendering slabs

27
System Performances

• Pure VolumePro
• 0.5 FPS
• Slab images cached in host memory
• 10 FPS
• Slab images cached in host and graphics memory
• 100 FPS (Without warping)

28
Performances(2)

• Slab Images Cached by both Host and Graphics
  Memory
• No Warping
• gt100 FPS
• Warped with mesh of grid size of 22 voxels
• gt50 FPS
• Warped with mesh of grid size of 11 voxel
• 10 15 FPS

29
Images
30
Properties of the Slab Structure

• Easy to create
• Empty space skipping.
• Occlusion culling
• Slab images can be reused
• Approximate perspective rendering with parallel
  projection.

31
Discussion and Future work

• With slab-based volume rendering, real time
  rendering of high resolution images can be
  achieved
• There are artifacts.
• Aliasing when switching slab sets
• Discontinuity between slab images

32
Acknowledgement

• NIH grant CA82402
• Naval Research grant N00014011034
• E-Z-EM Inc
• Viatronix Inc.
• University Hospital of Stony Brook.
• Min Wan, Baoquan Chen, Klaus Mueller, Manuel
  Oliveira and others
• Anonymous reviewers.

33
Thank You!