Process, WorkFlow in Medical Image Processing

1
Process-, Work-Flow in Medical Image Processing

• Guido Gerig
• http//na-mic.org

2
Need for Process Flow

• Image Processing and Analysis
• Sequence of processing steps (readers, filters,
  mappers, writers, visualization)
• Clinical studies between 30 and x00 datasets
• Research Prototyping Environment
• Process Flow System
• Fully automated (batch) and/or user-guided
• Guides user through processing steps
• Improved reliability and efficiency
• Relieves user from repetitive tasks
• Simplified sharing of processing sequences
• Process Flow System Beyond Script Files (?UNIX
  script/PERL/Python)

3
Example User-Guided 3-D Level-Set Segmentation
(SNAP)

• 3D Snake Segmentation
• Preprocessing (features)
• Initialization
• Post-editing
• User-guidance
• Challenge Use by non-experts
• Tool SNAP-ITK (Yushkevich, Ho, Gerig) 5years
  Project

4
Level Set Segmentation Pipeline

• Preprocessing
• Initialization
• Segmentation

A wizard guides the user through the segmentation
process
5
ITK-SNAP Tour Preprocessing
6
ITK-SNAP Tour Initialization

• Spherical bubbles or a coarse manual
  segmentation are used to initialize the level set

7
ITK-SNAP Tour Parameters

• Different user interfaces
• Intuitive mode
• Mathematical mode
• Preview of the forces acting on the level set

8
ITK-SNAP Tour Segmentation
9
Example EMS-ITK Atlas-based brain MRI
Segmentation
10
Example Hippocampus Shape Analysis Workflow
Manual Landmarking
Gray-value Normalization
MRI
Reformat
Hippocampus Segmentation via Model Deformation
Spherical Parameterization
SPHARM- PDM Shape
QC Shape Corresp.
Feature Computation e.g. Parcellation
or Difference to Model
Alignment Scaling
Prior Models
QC of Features Statistical Results
Statistical Analysis Of Features
11
Example DTI Analysis in large clinical study
(Ngt100)

• Co-registration of DTI
• Registration of DTI of each subject with
• structural MRI
• segmentation maps
• lobe parcellation
• user-defined ROIs
• Statistical analysis per ROI

Group 1
Group 2
12
DTI processing pipeline
4 DTI shots (.dcm)
dcm2hdr
4 DTI shots (.hdr)
DTIChecker
Average DTI (.gipl)
gipl2GE
Average DTI (GE format)
TensorCalc
FA/ADC maps (Gipl)
Tensor field
ROI and Lobe analysis
Fiber Tracking analysis
Analysis using Imagine
Using the FiberTracking tool
13
DTI processing pipeline (ctd.)
FA/ADC maps
Co-registration
sMRI (T1/T2/PD) EM-Segmentation ROIs
Brain Lobe Atlas MRI atlas template
Data Fusion Linear and nonlinear registration
ROI and Lobe Analysis
Writing Statistics
14
UNC Solution IMAGINE(Matthieu Jomier)
Download http//www.ia.unc.edu/dev
15
UNC IMAGINE

• Cross-platform
• GUI-based visual programming environment
• Command line applications integration Add your
  own modules
• Full integration ITK/vtk
• Modules executed as thread
• Memory manager allocate/disallocate mem.
• Visual feedback/log file
• Generates Source code (C) and makefile
  (Dyoxygen document.)
• Generates stand-alone cross-platform software
  with GUI

Imagine can generate Graphic User Interface
automatically. Here, an example demonstrating the
GUI generation for a recursive Gaussian filter.
16
Imagine Batchmake(Matthieu Julien Jomier)

• Parallel processing with BatchMake interface and
  script generation. With Batchmake, you can follow
  progress of your pipeline online

17
Demonstration Imagine 2

• Toy Example Data Fusion
• Registration of DTI to sMRI
• Registration T1 and T2/PD
• Registration of baseline DTI-0 to T2 (linear,
  nonlinear)
• Use transformation to register FA/ADC to T1/T2/PD

18
Discussion

• Process Flow Architecture significantly improves
  efficiency of research / exchange / time to
  market / large-scale studies
• Experience at UNC Since introduction in 04,
  the ITK-based ProcessFlow environment has become
  standard tool (backbone)
• NA-MIC Four uses
• Process flow in dedicated tasks (level-set
  segmentation, DTI processing, shape analysis,
  segmentation, etc.)
• Research environment to facilitate prototyping/
  exchange/ comparison Facilitates transfer of
  research tools to Core 2
• Clinical studies Core 3
• Process flow systems to set-up a proc. system for
  individual tasks
• Run Batch jobs on large clinical studies ?
  parallel/grid computing
• Verify results via qualitative visualization
• Training/Dissemination Core 5 Process flow
  systems with visual feedback are excellent for
  teaching of methodology and tools
• Architectures
• LONI Pipeline / AVS / SCIRun / UNC Imagine-1 and
  2 / MevisLab / .

19
Criteria

• ITK- and NA-MIC toolkit users dont need to
  program, does not require advanced programming
  skills
• Cross-platform
• Pipeline processing and visual programming
  environment
• Easy integration, e.g. command-line integration
  of own modules
• Facilitates tests/comparison/exchange even of
  complex software and whole systems
• GUI generation, e.g. creation of stand-alone
  cross-platform software from Pipeline
• Parallel Processing / Script Generation
• Clinical studies Multi-data processing
• Desirable for clinical studies Visual
  programming language structures like for loop,
  if then else and do while functions