2009 Winter Project Week:LesionSegmentationEfficiency

From NAMIC Wiki
Revision as of 22:55, 8 January 2009 by Mscully (talk | contribs) (→‎Key Investigators)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search
Home < 2009 Winter Project Week:LesionSegmentationEfficiency



Key Investigators

Mark Scully, MIND

Jeremy Bockholt, MIND

Steve Piper

Marco Ruiz

Luis Ibanez

Objective

Improve the running time and memory use of the new lesion segmentation method as the current module takes 90-120 minutes and up to 2 Gb of RAM.

Approach, Plan

Review code to determine custom pieces that can be replaced with existing filters. Make the custom pieces that can't be replaced multi-thread capable. Locate inefficient memory use and improve it.

See what if anything could be made to work in GWE (Marcus)?

Progress

  1. Luiz provided an itk statistics filter to replace current custom, single threaded, un-optimized code.
  2. Final piece on single threaded code in process of being replaced by a custom multi-threaded ITK filter.
  3. Prediction runtime reduced from between 1.5 - 2 hours to less than 9 minutes on a quad core machine.
  4. Prediction Memory footprint reduced from 2.5GB to 1.1GB
  5. With Marco's help our pipeline is now running on Grid Wizard


References