Core 1b – Engineering Computational Platform - National Alliance ...

burnwholeInternet και Εφαρμογές Web

5 Φεβ 2013 (πριν από 4 χρόνια και 4 μήνες)

151 εμφανίσεις

NA
-
MIC

National Alliance for Medical Image Computing

http://na
-
mic.org

Core 1b


Engineering


Computational Platform


Jim Miller

GE Research

National Alliance for Medical Image Computing

http://na
-
mic.org

1

4

2

5

3

Core 1b


Engineering

5 Aims / 5 Platforms

Architecture


tools, operating
paradigms, reporting mechanisms,
integration points

End
-
user platform


interactive
methods and information
visualization for longitudinal
analysis, exploratory data analysis,
and translational research

Computational platform


stream
processing, cloud computing,
statistical analysis, informatics,
machine learning

Data management


non
-
imaging and
derived data, DICOM and cloud
services

Software engineering and software
quality


navigable timeline for
revision control, build, test,
documentation and release

National Alliance for Medical Image Computing

http://na
-
mic.org

Computational platform

The objective of the Engineering component of
Core 1 is to provide software tools and software
development processes to deploy innovative
technology to clinical researchers,
support the
scientific algorithm innovation of the Algorithm
scientists
, and to foster a community to produce
high quality software.

Computing architecture


Desktop


Grid computing


Stream computing (GPGPU)


Cloud computing


Analysis platform


Feature libraries


Multivariate clustering


Machine learning


Statistical inference


Regression analysis


Information visualization


Informatics



DBP needs


Pathology segmentation


Longitudinal analysis


Patient specific analysis


Population analysis


Algorithm needs


Exploratory methods


Longitudinal frameworks


Interactive frameworks


Statistics


Machine learning


National Alliance for Medical Image Computing

http://na
-
mic.org

ITKv4

ARRA Funded effort from the National Library of Medicine


Kickoff 6/2010


v4.0 released 12/2011


v4.0.1 SPIE


v4.2 June 2012



National Alliance for Medical Image Computing

http://na
-
mic.org

ITKv4 Team

ITKv4 Team


GE Research


Kitware Inc.


University of Pennsylvania


Harvard University


University of Iowa


CoSMo Software


Mayo Clinic


University of Utah

ITKv4 A2D2 Team


Georgetown University


University of Utah


University of North Carolina
Chapel Hill


The Ohio State University


Carnegie Mellon University


Harvard University


William and Mary


Old Dominion University


Kitware Inc.

National Alliance for Medical Image Computing

http://na
-
mic.org

ITKv4 Overview


Apache 2.0 License


“Patented” directory removed


New software process


Git, Gerrit, Jira, cdash@home, Testing Data


Deprecated compilers


Visual Studio 6, Visual Studio 7, Borland 5.5, SUN CC < 5.9,
SGI CC, MWORKS, Cygwin, GCC < 3.4


64 bit improvements

National Alliance for Medical Image Computing

http://na
-
mic.org

ITKv4
-

Modularization

ITKv3


monolithic

organization

ITKv4


modular

organization

National Alliance for Medical Image Computing

http://na
-
mic.org

ITKv4


Level sets


ITKv3 provides several flavors of level sets


Implemented in a general finite difference framework


Each level set equation encapsulated in a separate filter


ITKv4 provides a more general framework


Level set equation can be constructed term by term


Advection terms


Propagation terms


Region terms


Regularization terms


Multi
-
material


Level set stopping criterions

National Alliance for Medical Image Computing

http://na
-
mic.org

ITKv4
-

Registration


Unbiased registration support


Composite transforms


Multithreaded metrics


New and updated metrics


Neighborhood correlation


Mutual information


Point set metrics


Tensor metrics


Vector metrics


New and update transforms


Displacement, Bspline, Poly
-
affine, Diffeomorphic


New optimizers


Efficient for high dimensional transforms


Automated parameter initialization



National Alliance for Medical Image Computing

http://na
-
mic.org

ITKv4
-

FEM


Broader FEM support (not just for registration)


ITK conformant


Memory and reference management improvements


SpatialObjects for IO


Removed other FEM IO mechanisms


FEMObject constructs problem instead of the solver


More details at

http://www.itk.org/Wiki/Refactoring_itk::FEM_framework_
-
_V4

National Alliance for Medical Image Computing

http://na
-
mic.org

ITKv4


GPU
(SPIE release or v4.1)


Architecture for ITK filters to utilize GPUs


Lazy synchronization of data between CPU and GPU


Factory mechanism


Samples of various “classes” of algorithms


BinaryThreshold, Mean, DiscreteGaussian,
AnisotropicDiffusion, DemonsRegistration

Filter on CPU

Filter on GPU

Filter on GPU

Filter on CPU

GPU buffer

GPU buffer

GPU buffer

CPU buffer

CPU buffer

CPU buffer

National Alliance for Medical Image Computing

http://na
-
mic.org

ITKv4
-

SimpleITK


On
-
ramp to ITK


Templateless layer for C++


Function Paradigm, Object
Paradigm


but no Pipeline Paradigm


Wrapping


Python, Java, C#, Tcl, Lua,
Ruby, R


Access to most ITK
algorithms


Simple image viewing


National Alliance for Medical Image Computing

http://na
-
mic.org

ITKv4
-

Video


Cameras, Ultrasound, Fluoro, OCT, …


Video bridge


OpenCV, VXL


Video grabber


Ringbuffer


Video pipelines

National Alliance for Medical Image Computing

http://na
-
mic.org

ITKv4
-

Microscopy


Point spread function models


Deconvolution


Denoising


Color correction


Colocalization


IO

National Alliance for Medical Image Computing

http://na
-
mic.org

Nipype


Python scriptable engine


Local and distributed execution


Semantically uniform access to
analysis tools


SPM, FSL, FreeSurfer, AFNI, Brains,
Slicer, Camino


Example:


Huntington’s Disease DBP tying all
their tools through Nipype


Nipype as the documentation vehicle
for the processing chain


Slicer Execution Model (Command
Line Modules) accessible through
Nipype


National Alliance for Medical Image Computing

http://na
-
mic.org

Feature libraries


Enable exploratory methods and learning based methods


ITK filter library


Gabor features


Haralick features


Entropy features


Polynomial fits to time varying data


Histogram features

Entropy

Histogram

Haralick

National Alliance for Medical Image Computing

http://na
-
mic.org

1

4

2

5

3

Core 1b


Engineering

Computational platform

Architecture


tools, operating
paradigms, reporting mechanisms,
integration points

End
-
user platform


interactive
methods and information
visualization for longitudinal
analysis, exploratory data analysis,
and translational research

Computational platform


stream
processing, cloud computing,
statistical analysis, informatics,
machine learning

Data management


non
-
imaging and
derived data, DICOM and cloud
services

Software engineering and software
quality


navigable timeline for
revision control, build, test,
documentation and release