NA-MIC National Alliance for Medical Image Computing http

NA-MIC National Alliance for Medical Image Computing http: //na-mic. org Process-, Work-Flow in Medical Image Processing Guido Gerig http: //na-mic. org

Need for Process Flow • Image Processing and Analysis: – Sequence of processing steps (readers, filters, mappers, writers, visualization) – Clinical studies: between 30 and x 00 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) National Alliance for Medical Image Computing http: //na-mic. org

Example: User-Guided 3 -D Level. Set Segmentation (SNAP) • 3 D Snake Segmentation: – – Preprocessing (features) Initialization Post-editing User-guidance • Challenge: Use by nonexperts • Tool: SNAP-ITK (Yushkevich, Ho, Gerig) 5 years Project National Alliance for Medical Image Computing http: //na-mic. org

Level Set Segmentation Pipeline • Preprocessing • Initialization • Segmentation A wizard guides the user through the segmentation process National Alliance for Medical Image Computing http: //na-mic. org

ITK-SNAP Tour: Preprocessing 1 Region competition stopping criterion (thresholding) 0 1 Intensity edge stopping criterion 0 -1 National Alliance for Medical Image Computing http: //na-mic. org

ITK-SNAP Tour: Initialization • Spherical ‘bubbles’ or a coarse manual segmentation are used to initialize the level set National Alliance for Medical Image Computing http: //na-mic. org

ITK-SNAP Tour: Parameters • Different user interfaces: – Intuitive mode – Mathematical mode National Alliance • Medical Imageof the forces for Preview Computing http: //na-mic. org acting on the level set

ITK-SNAP Tour: Segmentation National Alliance for Medical Image Computing http: //na-mic. org

Example: EMS-ITK: Atlasbased brain MRI Segmentation T 1 National Alliance for Medical Image Computing http: //na-mic. org T 2 Tissue Cortex

Example: Hippocampus Shape Analysis Workflow Reformat MRI SPHARMPDM Shape Alignment & Scaling Gray-value Normalization Spherical Parameterization QC Shape & Corresp. QC of Features & Statistical Results National Alliance for Medical Image Computing http: //na-mic. org Manual Landmarking Hippocampus Segmentation via Model Deformation Feature Computation e. g. Parcellation or Difference to Model Statistical Analysis Of Features Prior Models

Example: DTI Analysis in large clinical study (N>100) • Co-registration of DTI • Registration of DTI of each subject with: • structural MRI • segmentation maps Group 1 • lobe parcellation • user-defined ROIs Group 2 National Alliance for Medical Image Computing http: //na-mic. org • Statistical analysis per ROI

DTI processing pipeline 4 DTI shots (. dcm) dcm 2 hdr 4 DTI shots (. hdr) DTIChecker Average DTI (. gipl) gipl 2 GE Average DTI (GE format) Tensor. Calc FA/ADC maps (Gipl) Tensor field ROI and Lobe analysis Fiber Tracking analysis Analysis using Imagine Using the Fiber. Tracking tool National Alliance for Medical Image Computing http: //na-mic. org

DTI processing pipeline (ctd. ) FA/ADC maps Co-registration Brain Lobe Atlas MRI atlas template Data Fusion Linear and nonlinear registration ROI and Lobe Analysis Writing Statistics National Alliance for Medical Image Computing http: //na-mic. org s. MRI (T 1/T 2/PD) EM-Segmentation ROIs

UNC Solution: IMAGINE (Matthieu Jomier) Download: http: //www. ia. unc. edu/dev National Alliance for Medical Image Computing http: //na-mic. org

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 crossplatform software with GUI National Alliance for Medical Image Computing http: //na-mic. org Imagine can generate Graphic User Interface automatically. Here, an example demonstrating the GUI generation for a recursive Gaussian filter.

“Imagine” & “Batchmake” (Matthieu & Julien Jomier) Parallel processing with Batch. Make interface and script generation. With Batchmake, you can follow progress of your pipeline online National Alliance for Medical Image Computing http: //na-mic. org

Demonstration Imagine 2 Toy Example: Data Fusion: • Registration of DTI to s. MRI: – Registration T 1 and T 2/PD – Registration of baseline DTI-0 to T 2 (linear, nonlinear) – Use transformation to register FA/ADC to T 1/T 2/PD National Alliance for Medical Image Computing http: //na-mic. org

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 Process. Flow environment has become standard tool (backbone) • NA-MIC: Four uses: 1. Process flow in dedicated tasks (level-set segmentation, DTI processing, shape analysis, segmentation, etc. ) 2. Research environment to facilitate prototyping/ exchange/ comparison: Facilitates transfer of research tools to Core 2 3. 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 4. 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 / Mevis. Lab / …. National Alliance for Medical Image Computing http: //na-mic. org

Criteria • ITK- and NA-MIC toolkit users don’t 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 National Alliance for Medical Image Computing http: //na-mic. org