Chair for Computer Aided Medical Procedures & Augmented Reality
Lehrstuhl für Informatikanwendungen in der Medizin & Augmented Reality

THIS WEBPAGE IS DEPRECATED - please visit our new website

SUSI - The Segmentation User Interface

SUSI 2.0 - The Next Segmentation

"Verpeilt compiled"

"No morning without warning"

Author Contact: Ben Glocker

SUSI is a segmentation tool mainly for medical image data. It is developed for easy integration of new algorithms and rapid testing of prototypes. The GUI is based on FLTK 1.6. All rendering is done by the OpenGL library.

Existing Features

  • Import images
    • Meta Image Format (.mhd + .raw)
    • Analyze Image Format (.hdr + .img)
    • DICOM (based on the OpenSource DICOM library GDCM)
  • Export images
    • Meta Image Format (.mhd + .raw)
    • Analyze Image Format (.hdr + .img)
    • DICOM (based on the OpenSource DICOM library GDCM)
  • Image Processing
    • Transformation
      • Downsampling
      • Resampling (in resolution and physical voxel size)
      • Warping (rigid transforation)
      • Flipping
      • Rotating
    • Filtering
      • Inverting
      • Smoothing
      • Gradient filters
    • Histogram
    • Data Type Conversion
  • Segmentation
    • Methods
      • Thresholding
      • Region growing
      • Clustering
      • Graph cuts
    • Pre/Post-processing
      • Region of interest
        • Cropping
        • Rejoining
      • Inverting
      • Labeling
      • Combining segmentations
      • Erosion / Dilation
      • Boundary extraction
      • Skeleton extraction
      • Distance transformation (Chamfer distance)
      • Intensity replacement
  • Visualization
    • Slice view
    • Surface rendering (Marching Cubes) with OpenInventor export
    • Volume rendering
  • Misc
    • PET Attenuation map generation from CT image
    • Motion tracking & compensation
    • Diameter tracking

Recently Added

  • Diameter tracking (07.09.2005)
  • DICOM export (02.09.05)
  • Deletetion of brushes / Copy & Paste (Graph Cuts) (01.09.05)
  • Data type conversion (01.08.05)
  • Motion tracking & compensation (31.08.05)
  • 3D gaussian smoothing (26.08.05)
  • Joining images takes dimensions, physical size and patient position into account (12.08.05)
  • Intensity shift by -1024 for unsigned short images (11.08.05)
  • Intensity replacement with HU Units for background voxels (05.08.05)
  • Analyze import from Big Endian systems (04.08.05)
  • Analyze Image Format for Import and Export (02.08.05)
  • Display of image information (02.08.05)
  • Information about the segmentation properties as dialog (size, volume, voxel numbers, etc) (02.08.05)
  • Cross-Hair in slice view (01.08.05)

Upcoming Features

  • Several logical operators for combining segmentations
  • Line profiles
  • Histogram operations
  • Filter settings dialog
  • Automatic dialog closing

Feature Requests

Bug reports

  • Open
    • Sometimes seg faults when closing the volume rendering dialog
    • Problems with exported OpenInventor models
    • Still bugs in ROI and rejoining (e.g. after downsampling)
    • Wrap-around effects in filters (gauss2d, mean2d, median2d, laplace2d, sobel2d, prewitt2d)
    • Resampling of single slices (2d,z=1) doesn't work
    • Downsampling takes only integers as parameters (rounds floats to the next lower integer)
  • Fixed
    • Bug in Analyze export of physical size of voxel size (22.09.05)
    • Crash when drawing below lower border (01.08.05)
    • When drawing points and drawing over border, points are drawn on the oppsite side (01.08.05)
    • Region growing wraps at image borders (31.07.05)

Tips & Hints

  • Improving Graph Cuts results
    • Segment your object by manually setting the seed points (brushs) in the Graph cuts dialog
    • Add the segmentation to the component image using value 9 (red color)
    • If the inital segmentation is not completely inside the object that should be segmented apply an erosion before adding to the component image
    • Apply a dilation on your initial segmentation
    • Add this to the component image using value 10 (green color)
    • Export the component image
    • Open again the Graph Cuts dialog and instead drawing manually seeds load the component image
  • Measuring diameters of long shaped segmented objects
    • Segment your object by satisfying mehtods
    • Generate the Distance Map and save it
    • Load the Distance Map as input image
    • Import your segmenation as output image
    • Extract the centerline with the Skeleton algorithm
    • Replace the voxel intensites of the output image with the values of the Distance Map
    • The values of centerline voxels are the distances to the object border (diameters)
  • Replacing background voxels by HU for Air
    After resampling an CT image (e.g. for registration) the added voxels have intensity values of zero. To replace these values by Houndsfield units for air (-1024) do the following steps
    • Resample or load a resampled image
    • Apply a region growing on the background voxels by setting upper and lower threshold to zero
    • Invert the segmentation
    • Choose "Replace Intensities" from the Segmentation menu, then choose "Input + HU Air"
    • Save the output image
  • Deleting objects in several steps
    coming soon


Edit | Attach | Refresh | Diffs | More | Revision r1.32 - 22 Sep 2005 - 18:46 - BenGlocker

Lehrstuhl für Computer Aided Medical Procedures & Augmented Reality    rss.gif