Summer ITKv4 ClinicalGroupMeetingNotes
Overlap and Similarity
- Level of integration questions.
- Bridges to 3rd party libs
- Distribution questions: How to build/package with 3rd party libs. How to distribute data.
ITKv4 Features
- GPU
- Registration framework
3D Real-Time Physics-Based Non-Rigid Registration for Image Guided Neurosurgery
The following is a rough pipeline of the method with proposed classes.
Contributions: New filters, classes. Application?
Inputs: a segmentation mask, a mesh
Outputs: deformation field, transformed image(s)
- FeaturePointSelection3dFilter: No dependencies. Plan to start implementation with this filter.
- BlockMatching3Dfilter: Similar to Penn FEM registration classes? Perhaps only need to implement a new metric? Plan to use the GPU infrastructure, but also have a non GPU version.
- PBMSolver: PETSc dependence
- ImageWarp: Already in ITK
What support is needed?
- CMake integration w/ PETSc and MPI. Build / distribution issues.
- Further discussion and collaboration with the FEM, registration, and GPU groups.
Gaps:
- Mesh generation. Tetmesh reader / converter? Use Biomesh3D and bridge to ITK?
- Self-updating transform object
- PETSc & MPI within an ITK filter?
Data:
- Sample dataset for testing.
Distribution: External module?
Lesion Sizing Toolkit
Contributions: Functioning toolkit and application. Possibly new ITK filters classes. Data.
- Already in ITK as an external module. Contract is to port to use ITKv4 and distribute.
- Using spatial objects as inputs and outputs
Inputs: DICOM
Outputs: Lesion volume measurements and segmentations.
What support is needed?
- Does ITK want a tighter integration of these classes, and in this same form? Does this cover more general concepts useful to other groups. e.g. Enhanced canny edge detection
Gaps:
- Representing measures as a concept in ITK
- Annotations / metadata
Data: 60 datasets. Chest CT scans 1mm resolution. 200mb each. MIDAS? Store as DICOM? Automatically download using CTest.
Distribution: Currently external module. More integration?
ITK Algorithms for Analyzing Time-Varying Shape with Application to Longitudinal Heart Modeling
Contributions: New ITK module (particle system), filters, classes.
Inputs: Segmentations
Outputs: Point sets
- Uses the core of an existing code base built from ITK: ShapeWorks NITRC Repository
- Port significant portions to ITKv4
- New ParticleSystem module. New ITK filter process objects.
- Procrustes registration
- Generic infrastructure for point-based surface representations
- Cross-sectional and longitudinal analysis
What support is needed?
- Logistics of integration and distribution, including data.
- Future: support for mesh representation, GPU
Data:
- 25 longitudinal cardiac DE-MRI (1.25mm in-plane, 2.5mm thick) with segmentations of the left atrium. 2-4 datapoints each (pre ablation, 3mo, 6mo, 1 year)
- Need IRB to release image data
Gaps:
- Multivariate stats: Bridge to R for complex statistical analysis without going to file system. Only implement what is needed for within ITK algorithms.
Distribution: ITK Module. Test applications. Integration with third party applications (ShapeWorks)
