ITK Release 4/Microscopy

Microscopy

Overview

Microscopy applications bring new challenges to ITK at many different levels

• Very Large datasets (TeraBytes)
  • Examples:
    • Brainbow dataset = 11000x15000x1000 x (RGB) = 500 Gigabytes for one 3D image.
    • In-toto imaging dataset = 1000x1000x200x1000x5 ~ 500 GB for one imaging session (one organ)
    • Data acquisition
      • Today 100Gb / day
      • Tomorrow 1 Tb / day
      • 3D Stacks of 250Mb (8 seconds to acquire, 900 to deconvolve)
      • J.J. Field Optics Express 2010.
• Temporal data (2D+t, 3D+t)
• Multiple Components
• Very large number of objects of interest per image

Technical Details

The toolkit needs

• More explicit representation of time and channels
• Management of multi-resolution datasets
• Management of multi-gigabyte datasets (TeraBytes being the standard to look for in 5 years)
• Richer set of classes for object representations

Reference Applications

• Collaborating Groups
• Reference Applications

Related Events

• Related Events

Tcons

August 12th 2010

Attendees

• Alex Gouaillard - Cosmo
• Sean Megason - Harvard
• Arnaud Gelas - Harvard
• Kishore Mosaliganti - Harvard
• Marc Niethammer -UNC
• Cory Quammen -UNC
• Russ Taylor - UNC
• Marcus Hanwell - Kitware
• Brad Lowekamp - NLM
• Luis Ibanez - Kitware
• Dominik Spinczyk - Silesian University
• Dhanannjay Deo - Kitware

Topics

Challenges

• Large Images: Larger than local RAM
• Pixel Types:
  • Scalar, Vector, Tensor,
• Time
• Many channels
  • sometimes they are packaged as "RGB" for convenience, but are really independent channels.
• Cropping M-Dimensional image into N-Dimensional image (e.g. 4D --> 3D)
• Dealing with multi-resolution images
  • Jpeg 2000
  • TIFFs ?
• Non-regular spacing
  • Across Z (in optical confocal microscopy)
  • Across time (time points not acquired regularly)

Technologies Needed

• Being able to extract subpieces from a larger image
  • Boost hash map (MultiIndexContainer http://www.boost.org/doc/libs/1_39_0/libs/multi_index/doc/index.html)
• Need time-date-stamps in images (maybe at the slice level)
• Data structures
  • For image of multiple channels ?
    • Most common algorithms access a full channel at a time
      • instead of accessing all the components of a given pixel
  • Streaming? (off-memory)
    • Extend many filters to support streaming
  • Sub-sampled version of the image
• JAVA World
  • How to interface with them
  • Using Java libraries from C++
  • Using C++ libraries from Java
• File Formats
  • Synergy with Bioformats
  • Support Microscopy formats.
  • Reverse engineering existing code
  • Extend the ITK TIFF reader to support mult-channels (Luis volunteered to do it)
  • Synergy with ImageJ
    • ITK Plugins
• I/O
  • Reading specific channels (channel regions)
  • Reading specific time points or time regions
  • Reading subsampled versions of the image
  • Making them non-blocking
    • Attempt prediction of next region to load (when doing streaming).
• Remote computing, HPC
  • Large datasets
    • processing data remotely
    • sending scripts & to be executed in data servers

Meetings

• "Automated Imaging & High-Throughput Phenotyping" (December 5th-8th)
  • http://meetings.cshl.edu/meetings/autoimag10.shtml