Unified Roi Model
-- Ilya, Curtis, Josh'May 15, 2006
The Unified ROI Model tries to combine the continuous box model of the Flat Bounding Box(FBB) model as well as the discrete ROI5D model. One benefit of this is that each model concentrates on its strength. For ROI5D, the addition of a bounding box may prove a performance optimization for some queries. FBB gains a more exact discrete description language for the expression of contained sections. Added to the model are the 8 mandala-types, representing all possible bounding boxes. The typing of the bounding boxes is useful for restricting the inputs and outputs of modules based on the dimensionality of the region of interest. For simple cases, it will be possible that a ROI be defined in either FBB or ROI5D model. It is our intention, however, to provide an API which provides isomorphic translations from one type to the other as necessary.
Currently open questions include
- Is it possible to talk about slices of an existing ROI? What's the best way of doing this? (Convention, a FK, a specialised type)
- Does the proposal provide too many options for the casual user?
- What use cases produce non-isomorphic representations?
- Should every ROI5D have a BoundingBox?
- A ROI5D can handle 2D continuous shapes, but not higher dimensional ones. Is this a problem?
Moving forward
To begin work on the unified model, a preliminary review phase is scheduled for the week of the 19th of May, and a brief open discussion period until the 26th. After that, a process similar to that outlined on Roi5D Proposals will be followed, but with work done jointly by the various sites. This will include producing data matching each use case, implementing the model and the needed queries, and checking the expressivity (and lack of absurdity) in the model.
Comments / Suggestions / Brain-storming
Possible name changes
- BoundingBox --> ContinuousShape/OrthoCube (Jeremy) ContinuousROI (Josh)
- Roi5D --> HierarchicalSlicesROI (Jeremy) DiscreteROI (Josh)
