Flat Bounding Box

• Flat Bounding Box
  • Introduction
  • XML
  • FAQ
    • Currently the bounding box is restricted to 5-D. Why not extend it to N-D?
    • What is the distinction between Bounding Boxes & Shapes? Do the concepts have to be tied together, or can we treat them separately? Are Shapes the same as VisBio's Glyphs?
    • Why is a bounding box defined as continuous in XYZ, but potentially discontinuous in T & C?

Introduction

The heart of this model is a 5d bounding box. This bounding box is guaranteed to be continous in X,Y, and Z, but may be discontinuous across channels and timepoints. This bounding box is accompanied by an attribute from the "Shape" family of STs. The members of the shape family includes rectangles, binary masks, and 3D representations of surfaces. The "Shape" family is designed to be extended. 5D Bounding Boxes may be aggregated in a RegionUnion for purposes of tracking objects temporally, tracking cell lineage, etc. It is called "Flat" because there is little differentiation of types of BoundingBoxes, and the model does not allow BoundingBoxes to contain other BoundingBoxes.

XML

ST definitions

FAQ

Currently the bounding box is restricted to 5-D. Why not extend it to N-D?

The majority of our current use cases are 5-D: X, Y, Z, Timepoint, and Channel. We have one 6-D case, and that is spectral lifetime imaging. That is an outlier, and AFAIK, currently has no defined use case for a bounding box that extends across the spectral and lifetime dimensions. The reason not to expand the specification now is because we only have hypothetical use cases for bounding boxes with dimensionality greater than 5. In our past experience, expanding a spec to accomodate hypothetical use cases was wasted effort. We will expand the spec. later, as it becomes necessary.

What is the distinction between Bounding Boxes & Shapes? Do the concepts have to be tied together, or can we treat them separately? Are Shapes the same as VisBio's Glyphs?

A major function of this model is to specify volumes and their boundaries in up to 5 dimensions. Every finite volume can be approximated with a Bounding Boxes, that is, a hyper-rectangular region. A Bounding Box is insufficient to describe many volumes. For example, a sphere can be modelled mathematically, and attached to a bounding box. The bounding box is superfluous in this example, but is beneficial in that it:

• Provides simple criteria for obtaining original pixel data from the image server
• Allows normalized queries about overlapping regions. (e.g. Does an arbitrary volume overlap with this plane?)
• Allows software that does not support spheres to provide an approximation of the volume.

Spheres is one of many specifications for volumes, and we expect more specifications to be defined as developer and user needs progress. While the concept of Shapes could be separated from Bounding Boxes, I feel that the benefits of combining them outweigh any costs. Also, Bounding Box is an incomplete specification without Shapes. Shapes are potentially distinct from glyphs. Typically, the word glyph describes symbols that are used to mark points and/or convey meaning. The use cases to date use shapes to describe the boundaries of a volume. Glyphs such as the letter 'V' are not closed and do not describe a volume's boundaries. However, in other use cases, shapes could define an edge or a line to measure distance between two points. In summary, glyphs seem to be a subset of Shapes.

Why is a bounding box defined as continuous in XYZ, but potentially discontinuous in T & C?

Dimensions may be ordered or unordered. All our dimensions are represented by discrete sampling, but when dimensions are oversampled, they may be treated as continous. An extent of Start to End only has meaning on ordered dimensions. A discontinuous range makes the most sense on dimensions that are either unordered or undersampled. Modern microscopes oversample X and Y based on the size of molecules in relation to the wavelength of light. Z is also oversampled based on resolution. T is grossly undersampled based on rate of molecular activity. In most imaging applications, only X, Y, Z, and T are ordered dimensions. C is unordered in most imaging methods, including fluorescent images and RGB images of H&E histology slides. C is ordered only in spectral and lifetime imaging. C has been given extents to accomodate those specific imaging techniques. For the majority of imaging techniques, however, specifying an extents in C are In summary:

Dimension	Ordered Dimension	Oversampled and therefore Continuous
X	yes	yes
Y	yes	yes
Z	yes	yes
T	yes	no
C	maybe	no

An enumerated list of values is allowed for undersampled dimensions, because current use cases require that enumeration to be attached to a static spatial ROI. We have no use case where a static XY ROI needs to be attached to a non-contiguous set of Z-planes. All current use cases of 2d XY ROIs require the ability to adjust their position at each Z-plane, thereby not being static.