Introduction

Purpose

This document is a formal specification for the definition, storage and exchange of regions of interest (ROIs). This specification will be implementable in any programming language, and is intended to provide a common set of ROI types which will be usable in all image analysis software programs.

Scope

This specification defines abstract definitions of regions of interest, including details of how certain data structures and algorithms must be defined and behave, in order to ensure that ROIs work uniformly between the different programs and libraries implementing the specification. It also provides examples of serialised forms which may be used for storage and/or exchange. However, it does not define a file format; it is the responsibility of the implementors to integrate this model into their file formats as they see fit.

Reference implementation

This specification is accompanied by a reference implementation of the model. This implementation is intended to validate and test the correctness of the specification. It may be usable directly, however this is not the primary intention for its existence. Note that the reference implementation strives for complete correctness, and implementors of this specification may wish to provide additional optimisations to improve performance.

Concepts

• A ROI is an evaluation of a shape object
• A shape is defined by the rules which transform its representation (e.g. geometry, range within a dimension) into a a bitmask and/or greymask
• Each shape has a unique name (type) and number; the number is used for serialisation and versioning
• Each shape is described by one (or more) representations, these are the primitives which define the geometry or range within a dimension
• A shape object can be composed of one or more shapes, which can include transforms and shapes in arbitrary dimensions
• Each representation has a unique name and number; the number is used for serialisation and versioning
• Shapes which share representations may be freely interconverted; conversion is not required to be possible in both directions (e.g. square to rectangle or polyline to/from polygon)
• A shape is essentially a serialised expression which must be evaluated to create a usable ROI; given that certain shapes can contain other shapes, this provides for ROIs which are both extensible and of arbitrary complexity.
• All shapes can be serialised as a sequence of numbers
• Given that each shape can be reconstructed using its shape and representation numbers, which specify the exact sequence of numbers to deserialise to reconstruct the object, it is possible to exchange ROIs as simple text, or alternately as binary; more structured (but space inefficient) representations could be realised using XML.
• The object/function serialisation methodology used here is inspired by (but not derived from) the SSH FXP specification which defines the wire protocol for SFTP.