Printing Reconstructed Confocal Data on the 3D Printer

1
Printing Reconstructed Confocal Data on the 3D
Printer

• 08 November 2001
• Daniel E. Weber

2
Printing Confocal Data on the 3D Printer

• Transform serial section data (specifically Leica
  SP2 data)
• Identify relevant objects
• Calculate a 3D surface
• Make a solid print

3
Presentation Order

• Definitions
• Software packages needed
• Confocal data collection issues
• Clarifying the data
• Identifying segmenting the data
• Extracting the surface
• Converting to STL
• Printing on the GenisysXS

4
Definitions

• Segmentation
• Division of a dataset into identifiable portions
  corresponding to features of interest.
• Object Map
• A form of the dataset indicating to which
  segmented object each voxel belongs.
• Isosurface
• A surface in three dimensional space that
  corresponds to a constant data value.
• Volume Render
• A technique for visualizing a three dimensional
  dataset. Often confused with isosurface. No
  surface is calculated in a volume render.
• Voxel

5
Voxels
9.19
7.29

• Pixel Picture Element.
• Voxel Volume Element.
• 10 x 10 x 10 1000 voxels.
• Non-cubic voxels are a problem.
• Leica SP2 outputs a data file containing the
  voxel size.

7.29
3.12
7.29
9.19
9.19
7.29
uncorrected
9.19
7.29
corrected
6
Software Packages Required

• Photoshop 6
• Pre-processes the data
• Analyze
• Segments the volume
• Generates the surface
• Interchange or Rhino
• Converts DXF to STL
• AutoGen
• Prints the file
• All of these packages are available for Windows
  NT/2000 in the VMIL

7
Confocal Data Issues

• Evenly illuminated
• Evenly stained
• ROI in focus
• Sample present in all slices
• Save textual information about scope parameters
  for later use

8
Clarifying the Data

• Original Data
• One file per color channel
• Indexed color
• Enhancements
• Convert to RGB
• Auto-level
• Auto-contrast

?
9
Photoshop 6 Tools

• Use action set
• Confocal 3D Tools
• Channel -gt Auto RGB
• Use a batch script
• FileAutomateBatch
• Cycles through all images
• Inspect the files, if blurry may not be good data

?
10
Analyze Import/Export

• Can read TIF slices, but you must manually
  specify stacking order.
• Use the Import/Export Tool, sub tool Volume Tool.
• Select the files, edit the list if necessary, the
  save as file.vol.

11
Analyze Load As

• Load the dataset as a volume from the .vol file
  using LoadAs
• Adjust the voxel size.
• Note the change in dimensions. The data is now
  scaled and interpolated!
• Save as file.avw (for future reuse).

12
Analyze Save As

• Immediately save the dataset using SaveAs
• Save much time in future avoiding the resizing
  process.
• Always save frequently in Analyze.
• Processes are lengthy.
• Occasional crash.

13
Familiarize Yourself with the Data

• Volume Render Tool
• Initial view of the data
• Decide if worth pursuing
• Does not create a surface!

14
Image Segmentation Tools

• Isolating related voxels
• E.g. Skull, brains, tooth
• Often an art may require tweaking
• Dont want massive user intervention
• Analyze Tools
• Morphology Tool Step Editor
• Object Extractor
• Multi-spectral Analysis

15
Multi-spectral Image Analysis

• Uses color information to isolate related regions
• Color cube is divided up to create clusters of
  voxels that may be related
• Each voxels RGB value is evaluated, and a
  decision is made to associate that voxel with a
  particular classification
• User can set the tolerances

(0,255,255) cyan
(255,255,255) white
(255,0,255) magenta
(0,0,255) blue
(255,255,0) yellow
(0,255,0) green
(0,0,0) black
(255,0,0) red
16
Multi-spectral Algorithms

• Supervised
• Training file is created with human marked data
• Regions are selected that are known to be within
  each class
• Requires a minimum of two classes
• Requires marking on each slice (tedious)
• Very simple to get started
• User selects matching algorithm
• Nearest Neighbor
• Gaussian Cluster
• Neural Network
• Etc.

17
Multi-spectral Algorithms

• Unsupervised
• First voxel is put in a class by itself
• Tolerances are preset to determine subsequent
  voxel classification
• Number of classifications also specified
• Very fast
• May need further adjustment once segmented

18
Multi-spectral Object Maps

• Result of analysis object map.
• Map contains one 8-bit classification number per
  voxel.
• In the view window, the numbers are mapped to
  colors.
• Goal background and printable object data.
• Extra classifications may not hurt.
• Save the object map!

19
Defining the Surface

• ProcessModellingTiler
• GenerateParameters
• Threshold the data based upon object map
  classifications
• Adaptive deformation creates low-polygon count
  surfaces
• When done, save as a DXF file not as an STL
• May take 10-20 minutes

20
Converting to STL

• Printer requires STL format
• Rhinoceros or Interchange
• Rhino reposition into positive space
• Scale and reduce polygon count
• Save as binary STL

21
Loading into AutoGen

• ITG Tech Report 01-009 (Jo Wozniak)
• View as wireframe
• Scale to proper size
• Confirm printer is online
• Wait
• Wait some more
• Wait again
• Wait just a bit longer
• Remove part and clean

22
Additional Information

• Process works with RGB data, not just single
  color channel data.
• There have been a few tech reports on this
  process.
• New tech report geared towards confocal data
  coming soon.

23
Thanks to